OnPage.dev inside your AI assistant.

A 0 to 100 score for one page, the issues to fix first with why and how, AI readiness and the key facts: title, description, headings, words, speed and structured data. Client-side apps are scanned as rendered.

Up to 25 pages from the sitemap, or from the home page links when there is none, worked through step by step. Finds issues across the site, broken, orphan and duplicate pages, and powers link equity, internal links and edge rules.

Already crawled the site with Screaming Frog, Ahrefs or Semrush? Send up to 500 page rows and the internal links, and every audit tool works on the whole site, not just the first 25 pages.

Scans the next pages of a running audit on every call, so a whole site fits a free service. When it is complete you get the issues by code with the pages they affect, plus the average score.

Everything OnPage.dev measures on one page, by section: speed hints from the HTML, links and anchor texts, accessibility basics, image SEO, rich results and security headers. Use it when scan_page is not detailed enough.

Which AI crawlers may read the page per robots.txt (GPTBot, ClaudeBot, PerplexityBot, Google-Extended and more), whether llms.txt exists, 15 checks for being quoted in AI answers, and the page as a model reads it.

Takes the clicks, positions and revenue from your Search Console, GA4, Ahrefs or Semrush and ranks every fix by the traffic it can win. Also explains a drop: pass last period's clicks and decaying pages come first.

Your page next to up to 3 competitors that rank for the same search: score, AI readiness, words, headings, rich results and speed side by side, the facts only they give, and the fixes to catch up.

With citation counts from Ahrefs Brand Radar or a similar tool, scans the pages ChatGPT, Perplexity and Google AI cite and the ones they skip, and shows which checks the cited pages pass that yours do not.

Takes real AI answers from Ahrefs Brand Radar or another AI visibility tool and shows your share per engine, the questions where only competitors are named, what the AI searched, and the sites it cites to get onto.

Eight checks for one search term: title, meta description, H1, URL, the first 100 words, subheadings, image alt texts and keyword density, with a flag for stuffing over 3%. Each check comes with a tip.

Puts the page that should rank next to the pages of yours that outrank it, signal by signal: title, URL, headings, mentions, bold, italics, alt texts and internal anchors, with the plan to fix it.

Scans the pages that rank for a keyword and works out the page type Google rewards: guide, listicle, product, category, service page, forum or video. Then tells you whether your page matches, and what to change if not.

Groups your Search Console queries into topics and shows which page owns each one, where two pages compete for the same topic, which topics have no page yet, and which internal links to add.

Your assistant writes the sub-questions AI Mode and AI Overviews fan out for a topic. OnPage.dev rates each one answered, buried, partial or missing on your page and on up to 2 competitors, with the sections to add.

Internal PageRank and click depth from a site audit: pages the home page cannot reach, pages too many clicks deep, dead ends, strong pages that pass little on, and important pages that need more links.

Search terms from your Business Profile and map rankings per area become the service and location pages to write and the areas where you rank poorly.

A real browser on phone, tablet and desktop: is the H1 and call to action visible, does a cookie wall cover it. With screenshots. With screenshots, plus design checks such as low contrast and uneven spacing.

The page the way screen readers and browsing AI agents read it, from the accessibility tree: landmarks, the heading outline, and every link, button, image and field with the name it gets, including the ones announced as just "link".

How the structured data describes who and what the page is about, the way AI knowledge graphs read it: which entities there are, how they connect, dead profile links, and the JSON-LD that is missing.

Finds the passages on a page that best answer a question, the way AI answer engines pick text to quote, with their length and fit. Shows whether the answer is easy to lift out or buried in a long block.

Reading ease with the formula made for the page language, measured against the right target: the pages that rank for the keyword, or the norm for the page type. Lists the sentences that are hardest to read.

Checks the main text of a page in its own language for real errors only: spelling, grammar, confused words and compound words, with the sentence and the fix, leaving brand names and style advice out.

Every Open Graph and Twitter Card tag, and the share image itself: does it load, its real size in pixels, type and weight, against what Facebook, LinkedIn, X, WhatsApp and Slack need. Also a free REST API.

Opens the page in a real browser and captures the hits it really sends to Google: on a first visit, after accepting cookies and on a return visit. Finds double counting, tracking before consent, a banner that never updates Consent Mode and debug mode left on.

Core Web Vitals from real Chrome visitors, the field data Google uses: LCP, INP, CLS, FCP and TTFB on phone and desktop, with a pass or fail and the share of good visits. For the trend over time, use check_vitals_trend.

Reads your server access log lines and shows what Googlebot, Bingbot and AI crawlers really request: hits per bot, the errors and redirects they get, crawl budget lost on parameters, and sitemap pages Googlebot never visits.

A WCAG 2.1 AA report for the European Accessibility Act, every check mapped to its success criterion: language, alt texts, form labels, link names, headings and contrast, plus a list of what to test by hand.

Quick check of the cookie setup from the HTML: which consent platform runs, whether Consent Mode v2 is set to denied before the tags load, and whether the v2 signals are there. For live hits, use check_ga4_tracking.

Product structured data checked for Google: name, image, price, currency and availability, identifiers like SKU and GTIN, shipping and returns, variants, and the fields merchant listings need. Missing values come with the fix.

For a question people search, finds the section that answers it on the pages that rank and on yours, and compares the format: a paragraph and its length, a list and its items, or a table. Then tells you how to rewrite yours.

Valid language and region codes, x-default and the self-reference on a multilingual page, then every language version: does it load, is it indexable, does it link back, and do the canonical and html lang match. Writes the corrected tag set.

The trust signals Google and AI models look for across the site: About, Contact, Privacy and Terms pages, email, phone and address, company and VAT numbers, Organization markup and social profiles, and on articles the author and dates.

LocalBusiness markup, name, address and phone across the site, opening hours, geo and a page per location, and with your Google Business Profile: the same name, phone, address and hours on both, day by day.

Turns Search Console URL Inspection results and Pages or Enhancements report rows into causes and fixes: why a page is not indexed, and which template breaks stars, prices or FAQs in rich results.

For every type of Google manual action: what it means, the cleanup steps in order, the tools that check them, what to keep as proof, and a reconsideration request draft to fill in.

Groups the pages of a site audit by text similarity and shows which ones say almost the same thing, with their titles, words and canonicals. Per group the page to keep, and whether to merge with a 301, set a canonical or rewrite.

Classifies every query parameter in the site's links as tracking, session, search, sort, pagination or filter, and finds crawl traps: tracking in internal links, crawlable search results, filter combinations and pagination that canonicalises to page 1. Writes the robots.txt rules.

Fetches the page as Googlebot, GPTBot, ClaudeBot, PerplexityBot and more, next to a browser: firewall blocks, challenge pages, cloaking and bot-only redirects that robots.txt never shows.

Checks article templates for what Google Discover and News need: large image previews, a 1200 px image, a date, Article markup and a clear title, with your Discover rows.

Search Console clicks against GA4 sessions per page: a missing tag, consent loss, a redirect that drops the source, or mislabelled traffic.

Bing Webmaster crawl and index issues grouped with the fix, and the queries that do well in Google but barely in Bing.

The fixes for every issue: as HTML to paste, as a pull request, applied in WordPress, or served at the edge by a Cloudflare Worker on any CMS.

Turns a slow page into code, ordered by the Core Web Vital that fails: main image first, scripts deferred, early connections, font swap and image sizes, also as a Cloudflare Worker.

Sends the fixes to the OnPage.dev WordPress plugin. The site owner sees old and new value per field and approves with one click. The plugin writes them through Yoast, Rank Math or its own tags, and can undo every change.

Writes JSON-LD for Article, Product, FAQPage, Organization or BreadcrumbList from the page's own content and checks it against Google's rich result rules. Values it cannot read are marked TODO, so nothing is guessed.

Lists the images without alt text, your assistant looks at each one and writes it, then OnPage.dev checks the texts and ships them as HTML, a Cloudflare Worker, WordPress or code.

A ready-to-upload llms.txt built from the sitemap and home page: the site name, a short summary, and the main pages grouped by section with their titles. Check it afterwards with validate_llms_txt.

Writes robots.txt rules for 13 AI crawlers from a policy you choose: allow all, AI search only, block training, or block all. Merged into the site's current robots.txt, so your existing rules stay in place.

For a migration or a restructure: matches old URLs to new ones by path and title with a confidence score, lists the ones to check by hand, and writes the rules for _redirects, nginx, Apache or Next.js.

Sites that link to two or more competitors but not to you, ranked and typed with how to approach each, plus pages that mention your brand without a link, read live, with the sentence to quote in the email.

Shows how a title and meta description appear in Google: the pixel width against the cut-off on desktop and mobile, and the truncated text a searcher would see. Iterate on variants until they fit.

A writing brief for a search term from the pages that rank for it: the subtopics and questions they cover, the facts and figures they give, the length and the schema they use, and what your page could add.

Sentence and paragraph length, section size, question headings and how the site addresses its reader, learned from its best articles, so new drafts sound like the rest of the site.

Refreshes for queries almost on page 1 and new pages for missing topics, dated per week, steered by the formats that lifted traffic on this site.

Your agent writes, OnPage.dev checks: voice match with what to change, figures without a source and an SEO scan, before you approve.

Finds the audited pages that should link to a page, with the anchor text, and applies the links if you like: a Cloudflare Worker that links the first mention on each page, a WordPress plan or code edits.

One decision per page from a year of clicks, links and conversions: keep, refresh, merge with a 301, noindex or remove with a 410. Pages with value are never removed.

Scans HTML that is not live yet: a local build, a template or a draft from your editor. The same score and issues as scan_page, so your agent can check its own work before anything is deployed.

An SEO diff between two versions of a page, for example before and after a code change: the score change, issues fixed and introduced, and regressions like a stray noindex or a lost canonical, flagged before merging.

Compares a Vercel, Netlify or Cloudflare Pages preview with production, path by path: pages that break, titles or canonicals that vanish, noindex, lost structured data and canonicals that point to the preview host. Also runs as a GitHub Action on every preview.

Checks JSON-LD before you publish it: JSON syntax, @context and @type, full URLs, ISO dates, valid prices and currency codes, and Google's required and recommended fields for each rich result.

Checks a site's llms.txt against the llmstxt.org format: served as plain text, one H1 with the site name, a summary, sections with proper links, its size, and whether the links it lists actually load.

May Googlebot, GPTBot or ClaudeBot fetch this URL? Tests it against the live robots.txt with Google's own matching rules, and shows the exact user-agent group and rule that decides, so you know what to change.

Status codes and the full redirect chain for up to 20 URLs at once, and where each one ends up. Give the expected targets after a migration and every URL that lands somewhere else is flagged.

Checks every link from a page to other sites live: dead sources, cited pages that now redirect to a home page, moved pages and bot blocks, with an archived copy as a replacement for each dead one.

Scans a live page again and compares it with the previous scan: the score change, the issues that got fixed, new issues that appeared and the change in AI readiness. The proof that a deploy did what it should.

A small measurement script for your own browser tool, such as Claude in Chrome or Playwright, for staging sites, localhost and pages behind a login, where OnPage.dev cannot reach. No daily limit.

Runs the first-screen checks of render_page on the results from your own browser: is the H1 and call to action visible, pop-ups and cookie walls, tap targets, small text and what crawlers without JavaScript miss.

Finds the sitemap through robots.txt, follows the index and checks the format, the 50,000 URL limit, foreign hosts and lastmod dates that are all the same. Then checks a sample of URLs for redirects, errors, noindex and canonicals elsewhere, with corrected entries.

Builds sitemap.xml from a site audit with only the pages that belong in it: status 200, indexable and canonical to themselves, lists every page it left out and why, and never invents lastmod dates.

Custom extraction for up to 20 pages: a CSS selector, a regular expression or a JSON-LD path such as Product.offers.price. See which pages have the value, which miss it and how many distinct values there are, to verify a template fix at scale.

Google Sheets rows and a CSV link, a Slack message, one task per issue for Linear, Jira or Asana, and a checklist. Your agent sends it with the tools it already has.

Publishes a finished site audit as a clean report page with your agency name, ready to send as a link or save as a PDF: score, issues to fix first, sections and broken pages.

Scans a page again and publishes a shareable proof page on onpage.dev with the score before and after, the issues fixed and the ones still open. Made for clients. Not indexed, and it expires after 90 days.

Records every fix that shipped with its old and new value, the date and how it went out. OnPage.dev then keeps it live, measure_impact saves the result per fix, and get_fix_log turns it into a client report.

Tells Bing, Yandex, Naver, Seznam and Yep that the fixed pages changed, so they recrawl within minutes instead of days. The first call sets up the key file to upload. For Google, use URL Inspection in Search Console.

Takes broken and lost backlinks from Ahrefs or Semrush and the 404 hits from GA4, ranks dead pages by linking sites and visits, writes the 301s and link edits, and lists sites to ask for a link back.

Builds disavow.txt from your backlink data with only clear spam in it, puts anything with authority up for review, and warns you off uploading it without a manual action or real negative SEO.

From a site audit, one Cloudflare Worker rule per section: titles, descriptions, canonicals and social tags fixed on every matching page, with a preview. Only missing or out-of-range values change.

A Worker that serves GPTBot, ClaudeBot and PerplexityBot the rendered page from your own Cloudflare account, while visitors and Google get the normal one. It says so when a page does not need it.

Turns link prospects from your backlink tool into one short, personal email per site: the broken or missing link, your page and the anchor. They go to Gmail or Outlook as drafts once you approve.

Sorts fixes into a week-by-week plan for developer, content and outreach within the hours they have, quick wins first, and creates one ticket each in Linear, Jira, Asana or GitHub.

Checks that each changed page is indexable, gives the sitemap entries with a fresh lastmod, submits them to IndexNow and lists the pages to request in Search Console.

Drafts a short, personal reply for each Google review without one, unhappy customers first, with your contact for complaints. Posted through your Business Profile connector after you approve.

A daily check for regressions, with a private RSS feed and optional Slack or Discord alerts. Watch a competitor and every change they make comes with the counter-move for your page.

Audits the whole site every day, week or month, compares it with the last audit and posts what broke, turned noindex or got new issues to Slack, Discord or any webhook, without anyone asking.

Every change found on a watched page, newest first: the page going down or redirecting, a new noindex, blocked AI crawlers, a changed canonical or title, the score dropping, new and fixed issues, and the counter-moves for competitors.

Stops a watch and deletes its history, including the private RSS feed. Each connection can watch 3 pages at a time, so use it to swap a finished page for the next one you want to keep an eye on.

Every logged fix is checked about every 6 hours. When a deploy undoes one, the restore code goes to your webhook or, with the fix guard GitHub Action, into an issue in your repo.

Picks the home page and up to 10 key pages and checks them every day: score, AI readiness, issues with their codes, AI crawler access and Core Web Vitals per week. The history builds up for dashboards and client reports.

Compares a site audit with the previous one of the same site, kept 90 days: pages that broke, turned noindex or changed canonical, issues that are new or fixed per code, and pages added or gone.

Every 6 hours: robots.txt that blocks the site, a lost sitemap, noindex or a moved canonical on key pages, http without https. Alerts with the fix before rankings drop.

Reads competitors' sitemaps every week: new, updated and removed pages, flagged when they touch your topics.

Search Console, GA4 or Ahrefs numbers from before and after, against pages you did not touch, so seasonality is not counted as your win. With a fix log, the result is saved per fix.

Splits 20 or more pages of one template into two groups with equal traffic, gives a Worker that changes the title or description on one group only, and measures the result against the other.

Two periods from Search Console, compared page by page: lost rankings, lower search demand, fewer clicks at the same position because of an AI Overview, or pages that dropped out. Ranked by lost clicks, with the next tool per cause.

Daily Search Console or GA4 rows in, the days that broke from their normal weekday pattern out, with the pages behind each one and the Google core updates, spam updates and Search incidents that ran within 3 days.

The questions, comparisons and long conversational queries in your Search Console or Bing rows, per page, with the pages that rank but lose the click to an AI answer.

Results on page 1 whose click-through rate is far below normal for their position, with the clicks a better title and description could win. Quick wins that need no ranking change, tested with check_snippet.

Queries at positions 8 to 20 with real impressions, grouped per page, with the clicks a move into the top 5 could win and the pages that compete for the same query. The fastest ranking wins first.

Monthly clicks per page from Search Console, Ahrefs or GA4 become a refresh queue: pages that peaked and kept falling, with the clicks lost per month, and seasonal dips left out.

Core Web Vitals from real Chrome users, weekly for about 9 months, on phone and desktop. Give the date a speed fix went live and see the weeks before and after, and since when each metric passes.

Search Console plus your Ahrefs or Semrush tracker, with the keywords picked for you: pages you worked on, money terms, near wins and top 3. Each run lists what entered, dropped or switched page, and the next tool.

How often AI assistants cited your pages before and after the fixes, per engine, against pages you did not change, so a general rise in AI traffic is not counted as your result.

The feed as clean tables: overview, pages, issues over time, open issues, vitals, AI access and fixes. Dates and full URLs, so they join with GA4 and Search Console or their BigQuery exports in Claude Dashboards.

A complete blueprint for a live dashboard in Claude: realtime visitors from your own GA4, traffic from Search Console, and SEO health, fixes, speed and AI visibility from OnPage.dev, with ready BigQuery SQL and the prompt.

ChatGPT, Perplexity, Gemini, Copilot and Claude visits from your GA4: per assistant and landing page, and how they convert against other visitors.

Shows whether growth comes from SEO or from people searching your name: clicks, CTR and position for both, and how each side moved.

Your share of the clicks for a keyword set against competitors, weighted by position, with the keywords where a rival takes the clicks you could get.

Google Ads terms next to Search Console: where you rank top 3 and still pay, and which converting ad terms need a page.

One memory per site, across chats and clients: every OnPage.dev tool you call records what it found, shipped or measured. Open it once and the agent knows the history.

Measured results are tied to the fixes before them, so fix packs and priorities say what moved traffic on this site. Next steps follow: unmeasured fixes, open alerts, fixes worth rolling out.

A drop, alert or regression comes with the fixes, deploys and decisions of the last three weeks, and the whole timeline per page or kind.

Notes like "no prices in titles" or "migrated on 3 October" that every later brief and next step takes into account.

Behind a secret memory_key only you hold, deleted in one call or after 400 days without use. Opt in to share measured results anonymously (only the fix type and the rounded change) and learn what worked across sites.