AI detectors ranked by stated data retention
AI or Not states the shortest retention of submitted content in the index, none at all: "uploaded media are processed and deleted immediately after inference" (docs.aiornot.com, read 9 November 2026), and it ranked first among the cards at zero on the tie rule until 10 November 2026, when GPTKit joined at zero with two study records and took first place; ten cards now state a zero. Scribbr, ranked first until the third wave of cards on 9 November 2026, states the same for text: "we do not store or share your data. Submissions will never be added to our content database" (scribbr.com, read 8 November 2026). Twenty-one of the 72 detectors in the index state a retention period or state that submissions are not stored, and those twenty-one are ranked here (GPTKit, Fast-DetectGPT, Paperpal and AI Detector Pro joined on 10 November 2026, then Ircam Amplify and StealthWriter at 0 days and Trinka at 90 days later the same day, then McAfee Deepfake Detector at 0 days with the eleventh wave); 67 cards carry a retention statement of some kind, and the forty-three whose policies state no period, or two conflicting periods, are quoted in a second table in their own words (four of them, Illuminarty, Pindrop Pulse, Resemble Detect and Codeleaks, added on 9 November 2026). The fourth wave of cards on 9 November 2026 added Sightengine and Smodin at zero, TruthScan at 30 days and Compilatio at five years. Policies were read on 8 and 9 November 2026 from the privacy policy, documentation or product page each card cites.
Updated Policies read 8 and 9 November 2026
Short answer
Ten vendors say pasted content is not kept at all: GPTKit, AI or Not, Fast-DetectGPT, Scribbr, Sightengine, Undetectable AI, Paperpal, Phrasly, Smodin and WasItAI, the three new zeros added on 10 November 2026; AI Detector Pro states 90 days by its dated policy. SynthID keeps a digital signature of each uploaded file for 24 hours; Sapling caches a scored text for 3 days and offers a zero-retention flag on its API; Hive's default is 14 days, adjustable on request; TruthScan keeps submitted content 30 days by default; Compilatio stores submitted documents for 5 years from the end of the contract; Grammarly Authorship keeps a document's process record on its servers for up to 10 years, the longest stated period in the index. The other policies that mention submissions keep them for as long as the account is active, for as long as "reasonably necessary", or inside a report, and none of those states a number of days, so none can be ranked. A stated period is a promise in a policy, not a measured fact, and Phrasly's terms grant the company a perpetual licence over contributions in the same breath as its detector page promises immediate deletion. Policies read 8 and 9 November 2026. No detector score should be the only evidence in a misconduct or hiring decision.
Which AI detectors state the shortest retention of pasted text?
| Rank | Detector | Stated retention | Clause and source, with read date | Best for | Main limitation |
|---|---|---|---|---|---|
| 1 | GPTKit | 0 days (not stored) | "Once the detection process is complete, the content is immediately removed from our servers" (gptkit.ai FAQ, read 9 November 2026) | Readers who want a deletion-on-processing statement from a tool with 0 of 114 false positives | Personal data is transferred to India for processing; the privacy policy keeps it "only for as long as is necessary" with no period |
| 2 | AI or Not | 0 days (not stored) | "Privacy first: uploaded media are processed and deleted immediately after inference" (docs.aiornot.com, read 9 November 2026); free text tool: "No data stored" | Image, voice and video checks with no stored copy | Account data kept while the account is open; Live Voice logs kept up to 31 days; the one study record scored images |
| 3 | Fast-DetectGPT | 0 days (not stored) | "Detection text is only temporarily stored during processing and deleted immediately after completion" (fastdetect.net terms of use, served inside the demo's script bundle, read 10 November 2026); the repository runs locally with no upload | Researchers running the code locally, where nothing leaves the machine | The statement is the hosted demo's, not the paper's or the repository's; the demo's terms fall under the laws of the People's Republic of China |
| 4 | Scribbr | 0 days (not stored) | "we do not store or share your data. Submissions will never be added to our content database" (scribbr.com, read 8 November 2026) | Students checking drafts without an account | The statement sits on the detector page, not in the privacy policy; 1 study record |
| 5 | Sightengine | 0 days (not stored) | "Your content is deleted once it has been processed, unless you ask us to keep it or the law requires us to" (sightengine.com, read 9 November 2026) | API image, video and audio checks with deletion on processing | Demo files are kept as logs "deleted automatically after a limited period"; Feedback API content is kept to improve models; no study record |
| 6 | Undetectable AI | 0 days (not stored) | "Your text is never stored or used to train models" (undetectable.ai, detector FAQ, read 9 November 2026) | Free text checks with a stated no-storage line | The privacy policy of 18 March 2023 does not mention submitted text; the same company sells a humanizer; no study record |
| 7 | Ircam Amplify | 0 days (not stored) | "Audio files submitted to the Service [...] are automatically deleted or rendered inaccessible after completion of the analysis" (AIMD terms, version 2026.02, article 6.4, read 10 November 2026) | Labels and distributors checking music catalogues under a business contract | The clause is qualified by "unless otherwise expressly agreed in the Applicable Contract"; service metrics are retained; access is by application |
| 8 | McAfee Deepfake Detector | 0 days (not stored) | "will run on your device and will not store any audio or video data" (mcafee.com Privacy Notice, Last Updated January 2026, read 10 November 2026); product page: "McAfee doesn't store any of this audio or browsing history" | Readers who want a written no-storage line for an on-device voice check | The clause covers the checked audio and video; McAfee's wider notice keeps other personal data with no period; no study record |
| 9 | Paperpal | 0 days (not stored) | "Your sensitive academic work stays secure with Paperpal. We don't store your drafts or use them to train our AI models" (paperpal.com/tools/ai-detector, read 10 November 2026) | Researchers checking an unpublished manuscript | The Data Security FAQ keeps documents created in Paperpal until the user deletes them and Journal Fit uploads for 90 days |
| 10 | Phrasly | 0 days (not stored) | "processed securely and deleted immediately after analysis. We don't store your writing" (phrasly.ai, read 8 November 2026) | Writers who want no copy kept | Terms grant an "irrevocable, perpetual license" over Contributions (terms, read 8 November 2026); no study record |
| 11 | Smodin | 0 days (not stored) | "Your text is never stored or shared" (smodin.io, read 9 November 2026) | Free text checks with a stated no-storage line | Terms grant a "perpetual, irrevocable license" over User Content; the privacy policy does not mention submitted text; no study record |
| 12 | StealthWriter | 0 days (not stored) | "Text you submit for AI processing (humanization or detection). We do not store this text" (stealthwriter.ai/privacy, last updated 5 September 2026, read 10 November 2026) | Readers who want a written no-storage line from a no-account demo | The same company sells a humanizer; no accuracy figure and no study record |
| 13 | WasItAI | 0 days (not stored) | "We do not store submitted images beyond the duration required to complete the analysis request" (wasitai.com, read 9 November 2026) | Image checks with no stored copy | Log data kept 90 days; the detector carries a BETA label; no study record |
| 14 | SynthID | 1 day | "This signature is mapped to your user identifier and permanently deleted after 24 hours"; uploaded files "are immediately deleted" once results are returned (synthid.com, read 9 November 2026) | Checking whether a file carries a Google watermark | Detects only SynthID-watermarked media; no public text check; no study record |
| 15 | Sapling | 3 days | "AI detection results are cached on the entire text for 3 days; send zero_retention: true to score a text without Sapling keeping a copy" (sapling.ai, read 8 November 2026) | Developers who can set the API flag | The web tool's policy says content "is stored by us and used only to improve the Service" with no period |
| 16 | Hive | 14 days | "Our default policy is to retain customer data for 14 days. We can increase or decrease our retention time upon request" (docs.thehive.ai, read 8 November 2026) | API customers who negotiate a shorter period | Longer retention "may incur additional charges"; the company now trades as Deploy |
| 17 | TruthScan | 30 days | "Content Submitted for Analysis: Retained for 30 days by default, then deleted" (truthscan.com, read 9 November 2026) | Teams that want a written default with a zero-retention option | Zero Data Retention only on Business ($333 a month) and above, and the terms apply it to image and PDF submissions; no study record |
| 18 | AI Detector Pro | 90 days | "It is automatically deleted at the beginning of the 4th month after it was entered" (Privacy Policy, last updated 10 Jul 2023, aidetector.pro/pages/privacy, read 10 November 2026) | Readers who want a dated policy period for uploaded text | The Terms of Service say data "can only be expected to be maintained for 2 months"; the policy also keeps content until the user deletes it |
| 19 | Trinka | 90 days | "Premium user's processed data is automatically deleted from the server every 90 days" (trinka.ai/trust-center, read 10 November 2026); pricing table: "Data auto-deletion 90 days" on Basic and Premium | Researchers who want a dated default period with a real-time deletion tier above it | The privacy policy reserves the right to save documents and states no period; real-time deletion costs $500 a year |
| 20 | Compilatio | 1825 days | "duration of storage 5 years from the end of the contractual relationship" (support.compilatio.net, read 9 November 2026) | Institutions that need a stated period in a contract | The period runs from the end of the contract, so a live licence keeps documents longer; 1 study record |
| 21 | Grammarly Authorship | 3650 days | "The data remains on our servers for up to 10 years" once a report is opened; shared reports stay "until you delete your account" (support.grammarly.com, read 9 November 2026) | Writers who want a long-lived process record to show | A process record, not a classifier; report data deletable per document; no study record |
Policies that state no period
| Detector | What the policy says about submitted content | Source, read date |
|---|---|---|
| GPTZero | API: "We do not store or collect the documents passed into any calls to our API." Dashboard inputs are stored and "used in aggregate" to improve the service | gptzero.me, 8 November 2026 |
| Copyleaks | Data may be used "including by training Copyleaks' models"; direct customers may opt out via support; API content "is not stored, shared, or used to train public AI models" | copyleaks.com, 8 November 2026 |
| Originality.ai | Input text and results "may constitute scan history"; "Unless you opt out, we may use your scan history and scan results to help train and improve our models" | originality.ai, 8 November 2026 |
| Pangram | "This content is stored for as long as your account is active" and can be deleted from the history tab; zero data retention options for the enterprise API | pangram.com, 8 November 2026 |
| Winston AI | "We collect and store any text or images you submit"; deleting a report or the account deletes the content | gowinston.ai, 8 November 2026 |
| ZeroGPT | Content "is retained only for the period reasonably necessary to provide and operate the applicable Service"; not used to train models; no-retention modes for API and enterprise | zerogpt.com, 8 November 2026 |
| Turnitin | The AI result is "retained as part of the similarity report"; the institution chooses whether papers enter the repository; partial deletion of the AI component is not supported | guides.turnitin.com, 8 November 2026 |
| JustDone | Personal data stored "for as long as it is reasonably necessary"; "Your text is never sold or published" | justdone.com, 8 November 2026 |
| Proofademic | Submitted text processed automatically, "not used for model training or resale"; personal data kept "only as long as necessary"; no period for submitted text | proofademic.ai, 8 November 2026 |
| Walter Writes | Information retained "for as long as necessary to provide you with the Services"; text processed on self-hosted models; no period for submitted text | walterwrites.ai, 8 November 2026 |
| AIDetectPlus | Information "may be stored, collected, processed and used in countries other than the Republic of India"; deletion on request through a form | aidetectplus.com, 9 November 2026 |
| Codequiry | Code submitted without the Rapid add-on is added to a matching corpus and "retained indefinitely including after account closure"; Rapid submissions can be deleted on demand | codequiry.com, 9 November 2026 |
| Reality Defender | "We will not retain your personal information for longer than is reasonably necessary"; the factors are listed and no fixed period is stated | realitydefender.com, 9 November 2026 |
| Writer | "We will only retain your information for as long as we need it to fulfil the purposes for which we collected it"; inputs and outputs are never used to train models; the detector itself was sunset on 22 December 2025 | writer.com, 9 November 2026 |
| Turnitin Clarity | Writing Process Data (keystrokes, deletions, interactions) counts as Educational Data; retained "for the period necessary to enable the continued use of the Services", varying by type, then deleted | guides.turnitin.com, 9 November 2026 |
| BrandWell | "We retain information for as long as reasonably necessary for the purposes described in this Policy"; the Moxby listing says submitted text "is sent through the Moxby gateway to the analysis service" and warns against confidential text | brandwell.ai, 9 November 2026 |
| Surfer | Customer data processed "for the period necessary for the performance of the Contract" and up to 30 days after it ends for backups; entrusted data deleted or returned within 30 days of a request; nothing stated for text pasted into the free tool | surfer.ai, 9 November 2026 |
| Quetext | Detector FAQ: "Your scanned content is not stored, saved, or used to train any models"; privacy policy section 6.1: "We may include User Content and Service Usage Data in Quetext's AI Systems training" with an email opt-out, and personal data kept "only as long as necessary" | quetext.com, 9 November 2026 |
| Deepware | Terms: an uploaded sample "is immediately shared for review by the Service's Partners" under a worldwide, irrevocable licence; privacy policy: personal data kept "only for as long as is necessary"; no period for uploaded video | deepware.ai, 9 November 2026 |
| Sensity | Personal information stored "for as long as you use our Services or as necessary"; "We will delete your biometric data within three years of your last interaction", a clause the policy ties to selfie videos and identification documents, not to submitted media, so it is not ranked; submissions deletable through the API | sensity.ai, 9 November 2026 |
| Illuminarty | "Any image uploaded to our server will be kept only for the duration necessary to provide our service to you", a purpose-bound clause with no period and no deletion-after-processing statement (policy of 16 November 2022) | illuminarty.ai, 9 November 2026 |
| Pindrop Pulse | "Voice features", content-agnostic data points extracted from recordings, "are retained by Pindrop for a period of up to three (3) years and deleted thereafter"; the notice states no period for the recordings themselves, so the derived-record clause is quoted and not ranked | pindrop.com, 9 November 2026 |
| Resemble Detect | "Audio, image and video submitted for detection analysis is deleted after analysis unless it is flagged"; a flagged participant's recording is kept as the customer directs and no longer than three years; the clause states no period for submitted media as such, so it is quoted and not ranked | resemble.ai, 9 November 2026 |
| Codeleaks | Copyleaks' policy (24 August 2026): data may be used "including by training Copyleaks' models", direct customers may opt out through support; no period for submitted code | copyleaks.com, 9 November 2026 |
| PlagiarismCheck | Two documents conflict and no period is ranked: the AI detector page says "by default, we don't retain submitted documents, and educators can control their repository within integration"; the Privacy Policy (version dated 19.11.2025) says "Documents Checked: Until you delete them, or 1 year of inactivity" for individual users | plagiarismcheck.org/ai-detector and plagiarismcheck.org/privacy-policy, 9 November 2026 |
| Isgen | Two statements conflict and no period is ranked: the home page FAQ says "No, we do not store any data. Input history is maintained for users with a paid plan, but it can be deleted at any time"; the pricing page states "15 days Document History" (Academic) and "30 days Document History" (Ultimate) for stored paid-plan submissions | isgen.ai and isgen.ai/pricing, 9 November 2026 |
| Decopy | No period: the detector FAQ says "Decopy does not permanently save your content, and you can delete history records or uploaded documents anytime"; the Privacy Policy (last updated 24 March 2026) retains data "for as long as necessary" | decopy.ai/ai-detector, 9 November 2026 |
| Content Detector AI | No statement about submitted text: the archived Privacy Policy (11 May 2026) governs order information only | web.archive.org, 9 November 2026 |
| Detecting-AI.com | No period: the Privacy Policy (effective 10 October 2026) says anonymous submitted content "is processed to return the result and may remain briefly in necessary security or operational records" and signed-in scan history is kept in the account | detecting-ai.com/privacy-policy, 10 November 2026 |
| Binoculars | No policy: the paper and repository publish none, the code keeps text on the machine that runs it, and the Hugging Face Space runs under the platform's terms with a disclaimer that "the provided prediction is for demonstration purposes only" (the DetectGPT reading) | huggingface.co/spaces/tomg-group-umd/Binoculars, 10 November 2026 |
| Monica | No period: the Privacy Policy (last updated 14 June 2024) collects "the inputs, file uploads, and feedback (Content) you provide" for the History feature and retains personal data "only for as long as is necessary" | monica.im/privacy, 10 November 2026 |
| Wordtune | No period: the Privacy Policy (3 August 2025) says "Your Personal Data will be stored until we proactively delete it or you send a valid deletion request" and lists submitted text among collected content; the archived detector page carried no statement, and the tool is withdrawn | wordtune.com/privacy-policy, 10 November 2026 |
| Check For AI | No period: the archived Privacy Notice (1 January 2023) keeps information "for as long as it is necessary" and the Terms say transmitted Contributions "may be treated as non-confidential and non-proprietary"; the service was sunset in July 2023 | web.archive.org, 23 March 2023, 10 November 2026 |
| Writefull | Not ranked: the archived detector page (31 December 2024) said "Your data is encrypted and not stored", but the tool is discontinued and the live Privacy Notice (13 May 2025) keeps personal data "for as long as it is necessary" with no period for submitted text; the Writer treatment of an archived statement | web.archive.org, 31 December 2024 and writefull.com/privacy, 10 November 2026 |
| Passed.ai | No stated period: the Privacy Policy (29 January 2023) retains client personal information indefinitely and disposes of submitted third-party Content "within a commercially reasonable period of time, typically 90 to 120 days after it has been processed", a hedged range | passed.ai/privacy-policy, 10 November 2026 |
| ElevenLabs Classifier | No period for uploaded samples: the Privacy Policy (9 October 2026) keeps personal information "for as long as necessary" and "data we generate about your voice" for no longer than 3 years, a derived record (the Pindrop and Sensity reading); the FAQ says "We process your sample only to return a result" | elevenlabs.io/privacy-policy, 10 November 2026 |
| FakeCatcher | No product statement: the Intel Privacy Notice (27 March 2026) retains personal information "for the period necessary" with no figure, and the detector has no public submission route | intel.com privacy notice, 10 November 2026 |
| Deepfake-O-Meter | Not ranked: no privacy policy or terms page is published; the model groups API sets record_keeping_days 15 for a regular user, 30 for an advanced user and unlimited for a premium user, and the account page offers to delete the account and all associated data | model groups endpoint, 10 November 2026 |
| GPT Radar | No period: the archived Terms (22 January 2026) say the customer owns the Customer Data and GPT Radar "shall have the right to collect and analyze data" about use of the service, with a 30-day retrieval window after termination and no deletion period; the tool is withdrawn | archived terms, 22 January 2026, 10 November 2026 |
| Grammica | Not ranked: the live Privacy Policy says anonymous text is "Deleted immediately" and "automatically deleted from our systems within seconds to minutes", but the policy is dated January 2026, after the detector was removed from the site, and describes the relaunched editor; the archived detector page carried no statement | grammica.com/privacy-policy, 10 November 2026 |
| IvyPanda | No period: the Privacy Policy (25 October 2023) says "All User-submitted information at IvyPanda will be stored in our database" until the user asks by email for deletion | ivypanda.com/privacy-policy, 10 November 2026 |
| SEO.ai | No period: the Privacy Policy (10 April 2025, version 1.3) retains personal information "only as long as necessary to fulfill the purposes described above" and states nothing for pasted text | seo.ai/privacy-policy, 10 November 2026 |
| AI Writing Check | No period: the tool linked the Quill.org Privacy Policy, which retains personal data "only for as long as necessary to provide our services"; the tool has been withdrawn since August 2023 | quill.org/privacy, 10 November 2026 |
| RADAR | No policy: the paper, repository and model card publish none, the code keeps text on the machine that runs it, and the Hugging Face Space runs under the platform's terms (the Binoculars and DetectGPT reading) | Hugging Face Space, 10 November 2026 |
| Ghostbuster | Kept, no period: the README and demo state "all inputs to Ghostbuster are sent to the OpenAI API, and we also save the inputs for internal testing purposes" with no deletion period | github.com/vivek3141/ghostbuster, 10 November 2026 |
| Hiya Deepfake Detector | No period: the data policy linked from the Chrome Web Store (last reviewed 1 March 2024) processes personal data "only for so long as is necessary"; no statement covers the sampled audio | hiya.com data policy, 10 November 2026 |
GLTR, the GPT-2 Output Detector and DetectGPT run on the reader's own machine from open-source code, so no vendor policy applies to pasted text (open-source view); the cards record the repositories (GLTR, GPT-2 Output Detector, DetectGPT, read 9 November 2026).
One of ten rankings in the rankings hub, beside false positive rates and input limits. The business view and the admissions view weigh the same retention statements for procurement, and the home board shows every AI detector in the index on the same fourteen fields.
Which AI detectors should you check first?
- Ifyou paste student or client text and want no copy kept: Scribbr or Phrasly, both stating no storage, with Phrasly's terms read alongside.
- Ifyou check images and want nothing stored: WasItAI, with its 90-day log retention noted.
- Ifyou integrate an API and can pass a flag: Sapling, 3-day cache or zero retention per request.
- Ifyou need a negotiated, written period: Hive, 14 days by default.
- Ifyou check images, voice or video and want deletion after inference: AI or Not, or SynthID for Google-watermarked files, 24-hour signature only.
- Ifyou want the record kept, not deleted, to show later: Grammarly Authorship, up to 10 years.
- Ifyou check images or video through an API and want deletion on processing in the policy: Sightengine, with its demo-log clause noted.
- Ifyou paste text and want a no-storage line without an account: Smodin, with its terms' licence clause read alongside.
- Ifyou buy a business plan and want a written default you can shorten: TruthScan, 30 days by default, zero retention on Business and above.
- Ifyou are an institution that needs a stated period for a DPA: Compilatio, 5 years from the end of the contract.
What does each policy say, tool by tool?

AI or Not
- Verdict
- Fits readers checking images, voice or video who want nothing stored after the result.
- Value
- 0 days; "Privacy first: uploaded media are processed and deleted immediately after inference" (docs.aiornot.com, read 9 November 2026); the free text tool page states "No ads. No data stored."
- Best for
- One-off media checks on the free plan or through the API.
- Main limitation
- The privacy policy keeps personal data "for as long as you have an open account", Live Voice for Zoom keeps operational logs up to 31 days, and the one study record, Ha and Passananti 2024, scored images under the name Optic (arXiv 2402.03214, read 9 November 2026).

Scribbr
- Verdict
- Fits readers who want a plain "not stored" statement.
- Value
- 0 days; "we do not store or share your data. Submissions will never be added to our content database" (scribbr.com, read 8 November 2026).
- Best for
- Students and teachers checking drafts without an account.
- Main limitation
- The sentence sits on the detector page rather than in the privacy policy, and Scribbr has 1 study record, Walters 2023 (DOI 10.1515/opis-2022-0158).

Phrasly
- Verdict
- Fits writers who want deletion after analysis, read together with the terms.
- Value
- 0 days; "All text analyzed by our AI detector is processed securely and deleted immediately after analysis. We don't store your writing" (phrasly.ai, read 8 November 2026).
- Best for
- Writers checking their own drafts.
- Main limitation
- The terms, last updated 12 October 2026, grant Phrasly "an unrestricted, unlimited, irrevocable, perpetual license to use, copy, store and exploit your Contributions" (phrasly.ai/terms, read 8 November 2026); the privacy policy, dated 29 June 2023, does not mention submitted text. No study record.

Undetectable AI
- Verdict
- Fits readers who want a stated no-storage line on a free text check, read beside a thin privacy policy.
- Value
- 0 days; detector FAQ: "No. Your text is never stored or used to train models. For teams, our Business Solutions include even stricter data retention controls" (undetectable.ai, read 9 November 2026).
- Best for
- Writers checking drafts up to 10,000 words with no signup.
- Main limitation
- The privacy policy, last updated 18 March 2023, does not mention submitted text and keeps personal information "for as long as it is necessary" (undetectable.ai/privacy, read 9 November 2026); the same company sells a humanizer, and no study in the index tested the detector.

WasItAI
- Verdict
- Fits image checks where no stored copy is wanted.
- Value
- 0 days; "We do not store submitted images beyond the duration required to complete the analysis request" (wasitai.com, effective 27 March 2026, read 9 November 2026).
- Best for
- One-off image checks, 10 free per month.
- Main limitation
- Log data is deleted after 90 days and account data within 30 days of deletion; the detector is labelled BETA and no study record exists.

SynthID
- Verdict
- Fits readers checking a file for a Google watermark who want a stated deletion clock.
- Value
- 1 day; "Once the detection results are returned to you, the original media files are immediately deleted" and the anti-abuse digital signature "is mapped to your user identifier and permanently deleted after 24 hours" (synthid.com privacy notice, read 9 November 2026).
- Best for
- Verifying whether an image, video or audio clip came from a SynthID-integrated model.
- Main limitation
- "This is not a general AI detector"; it detects only media from companies that adopted SynthID, offers no public text check, and no study in the index tested it.

Sapling
- Verdict
- Fits developers who can pass the zero-retention flag.
- Value
- 3 days; "AI detection results are cached on the entire text for 3 days; send zero_retention: true to score a text without Sapling keeping a copy of it" (sapling.ai, read 8 November 2026).
- Best for
- Teams scoring documents through the API, with retention disabled per request or per team.
- Main limitation
- For the web tool the policy says "Your content is stored by us and used only to improve the Service" with no period; Walters 2023 returned uncertain on 31% of 42 human papers (DOI 10.1515/opis-2022-0158, read 8 November 2026).

Hive
- Verdict
- Fits API customers who want a written default period.
- Value
- 14 days; "Our default policy is to retain customer data for 14 days. We can increase or decrease our retention time upon request according to your needs (including not retaining your data at all)" (docs.thehive.ai, read 8 November 2026).
- Best for
- Enterprises negotiating retention in a contract.
- Main limitation
- "Increasing retention time may incur additional charges"; the brand changed to Deploy on 26 October 2026, and the one study record, Ha and Passananti 2024, scored images rather than text (arXiv 2402.03214, read 9 November 2026).

Grammarly Authorship
- Verdict
- Fits writers who want their drafting record kept to show an instructor or editor later.
- Value
- 3,650 days; "The data remains on our servers for up to 10 years" once a report is opened, and a shared report stays "until you delete your account" (support.grammarly.com, read 9 November 2026).
- Best for
- Students writing in Google Docs or Word who expect to be asked for process evidence.
- Main limitation
- Authorship is a process record, not a classifier, and returns no score for text it did not watch being written; report data is deletable per document; the record includes deleted passages and the names of websites visited while editing; no study record.

Sightengine
- Verdict
- Fits API customers who want deletion on processing written into the policy.
- Value
- 0 days; "The images, videos, text and audio you send to our API are only used to return results to you. Your content is deleted once it has been processed, unless you ask us to keep it or the law requires us to" (sightengine.com, last updated 6 November 2026, read 9 November 2026).
- Best for
- Platforms checking images and video at volume through the API.
- Main limitation
- Files tested in the web demo are kept as logs "deleted automatically after a limited period", and content sent through the Feedback API is kept "as long as it is useful to improve our models"; no study record.

Smodin
- Verdict
- Fits readers who want a plain no-storage line on a free text check, read beside the terms.
- Value
- 0 days; the detector page states "Your text is never stored or shared" and "Your content is encrypted and never stored", and the FAQ adds "We also never store your files" (smodin.io, read 9 November 2026).
- Best for
- Writers checking up to 50,000 characters without an account.
- Main limitation
- The Terms of Service grant Smodin "a worldwide, non-exclusive, royalty-free, perpetual, irrevocable license" over User Content, and the privacy policy of 30 March 2026 does not mention submitted text (smodin.io, read 9 November 2026); no study record.

TruthScan
- Verdict
- Fits teams that want a written default period and can pay for zero retention.
- Value
- 30 days; "Content Submitted for Analysis: Retained for 30 days by default, then deleted, unless retention is required for service improvement or otherwise agreed upon in a customer's service agreement" (truthscan.com, last updated 19 September 2025, read 9 November 2026).
- Best for
- Businesses checking images and PDFs under a service agreement.
- Main limitation
- The Zero Data Retention toggle is sold on Business at $333 a month and above, and the Terms of Service of 27 August 2026 say the setting "currently applies to image and PDF submissions" (truthscan.com, read 9 November 2026); no study record.

Compilatio
- Verdict
- Fits institutions that want a stated period written against the contract.
- Value
- 1,825 days; the privacy policy lists account data, including submitted documents, with a "duration of storage 5 years from the end of the contractual relationship", on servers in Chavanod, France (support.compilatio.net, read 9 November 2026).
- Best for
- Universities under GDPR that need a period a DPA can cite.
- Main limitation
- The five years run from the end of the contract, so documents under a live licence are kept for the licence plus five years; the GDPR page adds that submitted documents are not reused outside the institution's agreement.

GPTKit
- Verdict
- Fits readers who want a written deletion-after-processing line, the Sightengine reading.
- Value
- 0 days: "We temporarily store the data for processing purposes. Once the detection process is complete, the content is immediately removed from our servers" (gptkit.ai FAQ, read 9 November 2026).
- Best for
- One-off checks of text that should not sit on a server.
- Main limitation
- The privacy policy (effective 24 January 2023) transfers data to India and keeps Personal Data "only for as long as is necessary" with no period stated (gptkit.ai/privacy-policy, read 9 November 2026).
Fast-DetectGPT
- Verdict
- Fits researchers who run the open-source code locally and need no policy at all; the hosted demo states a zero.
- Value
- 0 days: the fastdetect.net terms of use read "Detection text is only temporarily stored during processing and deleted immediately after completion", text served inside the demo's JavaScript bundle rather than on a separate policy page (fastdetect.net, read 10 November 2026); the GitHub repository's local demo processes text on the user's own machine.
- Best for
- A local run from the repository, where submitted text is never transmitted.
- Main limitation
- The paper and repository publish no privacy policy; the zero is the hosted demo operator's statement, and that operator's terms are subject to the laws of the People's Republic of China (fastdetect.net, read 10 November 2026).

Paperpal
- Verdict
- Fits researchers who want a no-storage line on the detector page itself, the Scribbr and Smodin reading.
- Value
- 0 days: "We don't store your drafts or use them to train our AI models" (paperpal.com/tools/ai-detector, read 10 November 2026).
- Best for
- A manuscript check before journal submission.
- Main limitation
- The Data Security FAQ states other periods for other data: documents created in Paperpal are kept until the user deletes them and files uploaded for Journal Fit are deleted after 90 days (paperpal.com/data-security, read 10 November 2026); the zero governs detector drafts.

AI Detector Pro
- Verdict
- Fits readers who want a dated, stated deletion period rather than a promise of deletion on request.
- Value
- 90 days, the Quetext rule (the dated policy's period): "User content, which consists of all the text and documents that you upload, enter, or otherwise transmit when you use our products. We store this data in order to provide the core functionality to our users. It is automatically deleted at the beginning of the 4th month after it was entered" (aidetector.pro/pages/privacy, last updated 10 July 2023, read 10 November 2026).
- Best for
- Users who want a fixed horizon on stored scans.
- Main limitation
- Two documents give two periods: the Terms of Service section 34 says data "can only be expected to be maintained for 2 months" and the same policy says content is kept "until you delete them" (aidetector.pro/pages/terms, read 10 November 2026); the ranking takes the dated policy's stated period.

Ircam Amplify
- Verdict
- Fits business buyers who want a deletion-on-completion clause in a signed contract, the Paperpal and StealthWriter reading.
- Value
- 0 days: "Audio files submitted to the Service are processed solely for the purpose of performing the requested analytical output operations and, unless otherwise expressly agreed in the Applicable Contract, are automatically deleted or rendered inaccessible after completion of the analysis" (AIMD terms, version 2026.02, article 6.4, read 10 November 2026).
- Best for
- Catalogue scans where the audio must not sit on a vendor server.
- Main limitation
- The clause can be varied by contract, Ircam Amplify retains service metrics (volumes, scores, logs, model versions) for research and benchmarking, and the site's privacy policy link serves a transparency framework rather than a privacy policy (AIMD terms, version 2026.02, read 10 November 2026).

StealthWriter
- Verdict
- Fits readers who want a no-storage statement in a dated privacy policy, the Sightengine and GPTKit reading.
- Value
- 0 days: "Content submitted for AI processing (humanization or detection) is not stored at all. It is classified by a moderation tool and then either rejected or processed. We do not use submitted text to train models" (stealthwriter.ai/privacy, last updated 5 September 2026, read 10 November 2026).
- Best for
- A 500-word check without an account where the text should not persist.
- Main limitation
- The detector is the second tab of a humanizer product, the vendor publishes no accuracy figure and no study in the index tested it.

Trinka
- Verdict
- Fits readers who want a concrete default period from the vendor's documentation, the Hive reading.
- Value
- 90 days: the trust centre says "Premium user's processed data is automatically deleted from the server every 90 days" and the pricing comparison table prints "Data auto-deletion 90 days" for the Basic and Premium plans; the Confidential Data plan ($500 a year) states real-time deletion (trinka.ai/trust-center, trinka.ai/pricing, read 10 November 2026).
- Best for
- Manuscript checks on a plan with a stated horizon.
- Main limitation
- The privacy policy says Crimson AI "reserves the right to save documents on its systems according to its policy and Indian government regulations" and states no period, so the documentation's default is ranked and the policy quoted (trinka.ai/privacypolicy, read 10 November 2026).

McAfee Deepfake Detector
- Verdict
- Fits readers who want a no-storage statement in a dated privacy notice for a tool that runs on the device, the Hive and Copyleaks reading.
- Value
- 0 days: "The detection process will have access to any publicly available site from which you are playing the video input, but will run on your device and will not store any audio or video data" (mcafee.com Privacy Notice, Last Updated January 2026, read 10 November 2026); the product page adds "McAfee doesn't store any of this audio or browsing history" (mcafee.com/ai/deepfake-detector, read 10 November 2026).
- Best for
- A deepfake audio alert on a Windows AI PC where the clip must not leave the machine.
- Main limitation
- The clause governs the checked audio and video only; the same notice keeps other personal data with no stated period, the detector runs only on named PCs, and no study in the index tested it.
How we ranked
- Field sorted
- Days the vendor's policy says submitted text or images are kept, 0 where the policy says they are not stored or are deleted after analysis
- Direction
- Lower first
- Tie rule
- More study records first, then name
- Qualification
- A policy that states a period or states that submissions are not stored; "as long as necessary" or an open-ended account history is listed apart
- Qualified
- 21 of 72 detectors
Field sorted: the retention period for submitted text or images in days, as the vendor's policy states it, compiled on 9 November 2026. Direction: shorter ranks higher. Zero means the policy says submissions are not stored or are deleted after analysis. Tie rule: seven cards tie at 0, so ties break on study records per card, most first, then product name alphabetically (AI or Not and Scribbr 1 each, AI or Not first by name; Phrasly, Sightengine, Smodin, Undetectable AI and WasItAI 0, in name order). Qualification rule: a policy that states a number of days or states that submissions are not stored; a policy that keeps content "for as long as the account is active", "for as long as reasonably necessary" or inside a report states no period and is quoted in the second table, unranked. Where a vendor states one period for its API and none for its web tool, the stated period is ranked and the web tool's wording is given as the limitation. 5 of 22 qualified on the morning of 9 November 2026; 9 of 28 after the six cards of the third wave (AI or Not and Undetectable AI at 0, SynthID at 1 day, Grammarly Authorship at 3,650 days); 13 of 38 after the ten cards of the fourth wave the same day (Sightengine and Smodin at 0, TruthScan at 30 days, Compilatio at 1,825 days); 13 of 72 after the four cards of the fifth wave, none of which states a period for submitted media (Illuminarty, Pindrop Pulse, Resemble Detect and Codeleaks sit in the second table). Sensity's policy states one period, three years, for biometric data, which the policy ties to selfie videos and identification documents rather than submitted media, so the Sensity field is empty and the clause is quoted in the second table. Full rules: methodology page.
Why numbers differ between sources
A retention statement can live in four places and say four things. Phrasly's detector page promises deletion "immediately after analysis" while its terms claim a perpetual licence over contributions (phrasly.ai; terms, read 8 November 2026). GPTZero's FAQ separates API calls, not stored, from dashboard inputs, stored and used in aggregate (gptzero.me, read 8 November 2026). Copyleaks' pricing page describes a "Shared Data Hub, a free, secure library of millions of user-submitted documents" while its API FAQ says API content is not stored (copyleaks.com, read 8 November 2026). Codequiry's footer and policy say code without the Rapid add-on is kept indefinitely while its detector FAQ says submissions can be deleted at any time (codequiry.com, read 9 November 2026). The ranking takes the most specific statement about the detector's input and quotes the contradiction beside it.
A period says how long a vendor keeps text; a route says how a user removes it, and 17 cards state one, from a trash icon to a written request, on the deletion routes question, added 9 November 2026. The law firms view and the government view, added the same day, read the periods here beside deployment, and the compliance view beside each vendor's named standards.
Two twelfth-wave views read this field beside others: the customer support QA view requires a stated period with an API, a short-text statement and a bulk route, and the team seats and SSO view notes the data terms that enterprise and team tiers add, both read 10 November 2026.
Questions about AI detector data retention
Which AI detector keeps your text the shortest time?
AI or Not, Scribbr, Phrasly, Sightengine, Smodin, Undetectable AI and WasItAI state that submitted content is not stored at all, per their pages read 8 and 9 November 2026; AI or Not ranks first on the tie rule, 1 study record and the first name alphabetically among the two cards with a record. SynthID's 24-hour signature, Sapling's 3-day cache, Hive's 14-day default, TruthScan's 30-day default, Compilatio's 5 years and Grammarly Authorship's 10 years are the only numeric periods in the index.
Do AI detectors store the text you paste?
Thirty-four of the 72 cards state that submitted content is kept in some form or for some period (PlagiarismCheck and Isgen, added 10 November 2026, each publish two statements that conflict and are not ranked; Monica, Wordtune, Check For AI, Passed.ai, ElevenLabs and Deepfake-O-Meter, added later the same day, state a period for personal or tier data and none for the submission; GPT Radar, IvyPanda, SEO.ai, AI Writing Check and Hiya, added with the eleventh wave, do the same, and Grammica's live zero is dated after its detector was withdrawn), per the policies read 8 and 9 November 2026, with Grammarly Authorship stating the longest period, up to 10 years, and Writer and Turnitin Clarity stating no period: GPTZero stores dashboard inputs, Winston AI stores every submission until a report is deleted, Pangram keeps history while the account is active, and Copyleaks and Originality.ai may use submissions to train models unless the customer opts out. Seven say no, and three run on the reader's own machine. The question page on stored text quotes every clause in one table.
Which AI detectors train on submitted text?
Copyleaks ("including by training Copyleaks' models", opt out via support) and Originality.ai ("we may use your scan history and scan results to help train and improve our models", opt out available) say they may, per their privacy policies read 8 November 2026. ZeroGPT, Pangram, Phrasly, Proofademic and WasItAI state that they do not; GPTZero says dashboard data is used "in aggregate". Quetext's detector FAQ says scanned content is not used to train models while section 6.1 of its privacy policy, last updated 14 October 2026, says User Content may be included in training with an email opt-out (quetext.com, read 9 November 2026).
Does an API retention promise cover the web tool?
Not on its own. GPTZero does not store API inputs but does store dashboard inputs; Sapling caches API texts for 3 days unless zero_retention is set and says web content "is stored by us"; Copyleaks does not store API content but describes a shared library of user-submitted documents on its pricing page, all read 8 November 2026. A buyer of the web tool should read the web tool's clause.
Which detector keeps data the longest?
Codequiry's policy, last updated 22 October 2026, says code submitted without the Rapid add-on is added to a cross-account matching corpus and "retained indefinitely including after account closure" (codequiry.com, read 9 November 2026). No other policy in the index names an indefinite period; the longest stated period is Grammarly Authorship's "up to 10 years" for an opened report, followed by Compilatio's 5 years from the end of the contract (support.compilatio.net, read 9 November 2026) (support.grammarly.com, read 9 November 2026).
Is a stated retention period verified?
No. The index records what each policy says and when it was read; no vendor's storage was inspected. Hive's documentation states a 14-day default and Sapling's a 3-day cache, read 8 November 2026, and the ranking reports those statements as statements.
Does retention matter for FERPA, GDPR or a DPA?
The policies that name a framework are GPTZero ("Compliant with FERPA, GDPR, and CCPA"), ZeroGPT (infrastructure hosted in Germany under GDPR) and Reality Defender (US, UK and EU certifications listed in its FAQ), read 8 and 9 November 2026. A retention period is one term a data processing agreement would fix; the school districts view sets the same statements beside LMS integration.
How often is this ranking updated?
Policies are re-read on the monthly recheck, next on 8 December 2026, and the ranking changes whenever a policy adds or removes a period, with the Updated line matching. A new card enters on the day its retention field is read. Policies from the index, compiled 9 November 2026.
Sources and revisions
Sources
- Scribbr AI detectorscribbr.com/ai-detector
- Phrasly AI detectorphrasly.ai/ai-detector
- Phrasly terms and privacyphrasly.ai/terms/privacy
- WasItAI privacy policywasitai.com/privacy
- Sapling privacy policysapling.ai/privacy
- Hive documentation FAQdocs.thehive.ai/docs/frequently-asked-que...
- GPTZero privacy policygptzero.me/privacy-policy.html
- Copyleaks privacy policycopyleaks.com/privacy-policy
- Originality.ai privacy policyoriginality.ai/privacy-policy
- Pangram data privacypangram.com/data-privacy
- Winston AI privacy policygowinston.ai/privacy-policy
- ZeroGPT privacy policyzerogpt.com/privacy-policy
- Turnitin AI writing detection FAQsguides.turnitin.com/hc/en-us/articles/284...
- JustDone privacy policyjustdone.com/legal/privacy-policy
- Proofademic privacy policyproofademic.ai/privacy-policy
- Walter Writes privacy policywalterwrites.ai/privacy-policy
- AIDetectPlus privacy policyaidetectplus.com/privacy-policy
- Codequiry privacy policycodequiry.com/privacy-policy
- Reality Defender privacy policyrealitydefender.com/t-c/privacy-policy
- AI or Not API documentationdocs.aiornot.com
- Undetectable AI detector FAQundetectable.ai/ai-detector
- Undetectable AI privacy policyundetectable.ai/privacy
- SynthID Detector privacy noticesynthid.com
- Grammarly Authorship support articlesupport.grammarly.com/hc/en-us/articles/2...
- Writer privacy policywriter.com/legal/privacy
- Turnitin Services Privacy Policyguides.turnitin.com/hc/en-us/articles/273...
- Sightengine privacy policysightengine.com/policies/privacy
- Smodin AI detector page and FAQsmodin.io/ai-content-detector
- TruthScan privacy policytruthscan.com/privacy
- TruthScan pricing (Zero Data Retention)truthscan.com/pricing
- Compilatio privacy policysupport.compilatio.net/hc/en-us/articles/...
- BrandWell privacy policybrandwell.ai/privacy-policy
- Surfer privacy policy for customerssurfer.ai/legal/privacy-policy-customers
- Quetext privacy policyquetext.com/privacy-policy
- Deepware terms of servicedeepware.ai/us/terms-of-services/index.html
- Sensity privacy policysensity.ai/privacy-policy
- DetectGPT repositorygithub.com/eric-mitchell/detect-gpt
Revision log
- Ranking created, 5 of 22 qualify
- Six wave 3 cards added, ai or not, undetectable ai, synthid and grammarly authorship join, 9 of 28 qualify, writer and turnitin clarity added to the second table
- Ten wave 4 cards added, sightengine and smodin at 0 days, truthscan 30 days and compilatio 1,825 days join, 13 of 38 qualify
- Four wave 5 cards added, none ranked: illuminarty's purpose-bound clause, pindrop's voice-features clause and resemble detect's flagged-media clause state no period for submitted media (the sensity reading), and codeleaks states none
- Eleven wave 6 records added
- Eighth-wave pages linked
- Nine wave 9 cards added
- Twelve wave 10 cards added
- Self-publishing view linked
- Nine wave 11 cards added
- Twelfth-wave pages linked
). Illuminarty, one of four cards added in the same wave, joined with its one image record from Ha and Passananti 2024, so 18 of 72 cards now qualify.: register plate](/img/independently-tested-480.webp)


