nPage.dev Use in Claude & ChatGPT

Privacy

OnPage.dev is free to use without an account. We collect as little as possible, and this page says exactly what that is.

Who is responsible

OnPage.dev is an independent service based in the Netherlands. For any question about your data, email hi@onpage.dev.

What happens when you scan

Your IP address

Your IP address is used to limit how many scans can run per minute, so the service stays free and fast for everyone. For that it is held in memory for that minute and not stored.

Two features have a daily limit per visitor: real-browser checks (5 a day) and page watches (3 at a time). For those we store a one-way code made from your IP address with a secret key, never the address itself. The code cannot be turned back into your IP address. It is kept for 2 days for the browser checks, and for as long as a watch exists (at most 90 days) for watches.

Cookies and storage

We set no cookies and use no advertising or tracking. Your browser keeps two small items: your choice of light or dark mode (localStorage) and a technical flag that reloads the page once after an update (sessionStorage). Neither is sent to us.

Visitor statistics

We use Cloudflare Web Analytics to count visits. It works without cookies and does not build profiles of visitors.

Using OnPage.dev inside an AI assistant

When you connect OnPage.dev to Claude, ChatGPT or another assistant through our MCP server, the assistant sends us the address or HTML to scan and receives the results. We handle that request exactly like a scan on this site. A random, signed connection ID is used only for rate limiting and is not stored. What the assistant does with the results falls under the privacy terms of that assistant.

HTML you send to the scan_page · html tool is analysed in memory and not stored.

Site audits (site_audit · start, or crawl rows you send to site_audit · import) are kept for one hour so your assistant can work through them. When an audit finishes, a compact snapshot stays for 90 days so site_audit · compare can show what changed next time: per page the address, status code, score, title, canonical, whether it is noindex and the names of the issues found. It is stored under a scrambled (hashed) version of the site address. Page text is not kept.

Backlink, 404 and visit rows you pass to backlinks · broken or backlinks · opportunities, AI answers you pass to ai_visibility · gaps, and monthly clicks you pass to find_opportunities · content_decay, are used for that one answer and not stored. When backlinks · opportunities reads a page live, only whether it mentions and links to you is kept, for 15 minutes.

A rank tracker (track_rankings) stores, per run, the keywords it follows with their position, clicks, impressions and ranking page, plus the site address, the money keywords and webhook you give, for up to 400 days or until you stop it. It keeps the first run and the last 90. The rest of the Search Console, Ahrefs, Semrush or GA4 rows you pass is used for that run only.

The outbound link check (check_links · outbound) asks the Internet Archive (archive.org) for a saved copy of links that no longer work; only those link addresses are sent. Backlink rows you pass to backlinks · disavow are used for that one answer and not stored.

The spelling check (text_quality · spelling) sends the main text of the page you ask about, up to 10,000 characters, to LanguageTool (languagetool.org) to find errors. That is public text from the page itself; the result is kept for 15 minutes. Audit report pages from site_audit · report hold the summary of that audit (scores, issue names, example addresses) and the brand name you give, are public for anyone with the link, not indexed, and are deleted after 90 days.

A scheduled site audit (watch · site_audit) stores the site address, how often to audit, your webhook address if you give one, and the result of the last audit, until you stop it or for 400 days. Each audit leaves the same 90-day snapshot as a normal site audit. To limit schedules per visitor we use the same one-way code as for page watches.

A site memory (site_memory) is only created when you or your AI assistant ask for one. It stores the site address and, for every OnPage.dev tool called in a chat where the memory is open, the tool name, the page address, a one or two line summary of the result and a few numbers from it (for example a score, a count or the clicks you passed in), plus notes you add. Results from guards, scheduled audits, competitor watches and fix logs started in that chat are added too. It does not store the full results, your raw data rows or page content. Access needs the secret memory_key, which we store only as a one-way hash; we cannot recover it. The memory is deleted when you call site_memory · forget, or after 400 days without use. It is stored in Cloudflare D1.

The OnPage.dev WordPress plugin is only connected when a site administrator clicks Connect. We then store the site address, a site key and a secret (both only as one-way hashes), and the change sets an AI assistant sends for that site: the page address, the proposed values and, once the administrator decides, which changes were applied and their old values. Nothing is written to the site without the administrator's approval, and OnPage.dev never logs in to it. Disconnecting or deleting the plugin removes the site and its change sets; otherwise they are deleted after 400 days without use. They are stored in Cloudflare D1.

Sharing what worked is off unless you turn it on for a memory (share_learning). When it is on, every measured result in that memory adds one row per fix type to an anonymous pool: the fix type (for example title or meta description), the measured change rounded to a whole percent, and the month. It does not include the site, any address, your numbers or any text; each memory is represented by a one-way code so one site cannot dominate the results. Results from the pool are only shown once a fix type has at least 20 measurements from at least 10 different sites. Turning sharing off, or deleting the memory, removes all rows that memory added.

A site guard (watch · guard) stores the site address, up to 6 page addresses, your webhook address if you give one, and the result of the last check (status codes, noindex, canonical and title of those pages, and whether robots.txt blocks crawlers), until you stop it or for 400 days. A competitor content watch (watch · competitors) stores the competitor addresses, your topics, and the URL list and lastmod dates from their public sitemaps, for the same period. Both use the same one-way code to limit how many each visitor can run.

Traffic numbers your assistant passes to the prioritize_fixes tool, for example clicks, positions or keywords from Search Console, GA4, Ahrefs or Semrush, are used for that one answer and not stored. We never connect to those services ourselves.

Fix logs from the fix_log · log tool store the page addresses, the old and new values and the dates you log, plus whether each fix is still live, for 400 days after the last change. If your assistant runs measure · impact with a fix log, the before and after numbers and the change per logged page are saved in that log. A fix log can only be read with its id, which only your assistant receives. Email hi@onpage.dev with the id to have it deleted.

Who processes data for us

OnPage.dev runs on Cloudflare, which handles hosting, the network and the visitor statistics under its own data protection terms. We do not sell or share data with anyone else.

Your rights

Under the GDPR you can ask what we hold about you, and ask us to correct or delete it. Because we do not keep accounts or link scans to people, there is usually nothing to look up, but you can always email hi@onpage.dev. You can also file a complaint with the Dutch Data Protection Authority (Autoriteit Persoonsgegevens).

Changes

If this changes, we update this page and the date at the top.