Twenty-seven agents in the engine. What each one does today.
The audit is the front door, and it runs no model. It reads your pages with a crawler that names itself and obeys a robots.txt group that names it. Behind it is a registry of separate agents, each responsible for one discipline: 17 run in order in the full pipeline (Growth plan and up), 5 are not offered because they would act on your accounts or measure nothing, and 5 need an input a website cannot give. Here is every one that runs or is built; the five we do not offer are on Why us.
Every check we run, on one screen
52 checks that between them raise 106 distinct findings. Each square is one check; hover or focus it to read what it looks for. Nothing here is a category we invented for a screenshot: it is the list the engine iterates.
- Its relative links stay on this site
- It names a clean, absolute canonical address for itself
- Its canonical address answers directly
- It declares its text encoding where browsers look for it
- The site publishes robots.txt and names a sitemap in it
- Its language alternates are valid
- Each language alternate answers and links back
- Each language alternate can be read and points one way
- It is served compressed
- It is served over HTTPS
- Search engines may index it and follow its links
- Its links to its own pages use https
- Its links to its own pages are ones crawlers may follow
- Its address reaches the page in one secure hop
- Its robots directives are valid and consistent
- Your robots.txt allows this page
- Its links agree with its own crawl rules
- It carries structured data
- It loads nothing over plain http
- It sends the security headers browsers act on
- Its sitemap listing agrees with its indexability
- Its structured data actually parses
- Its address has no doubled slash or capital letters
- It has a viewport tag for phones
- Its viewport lets a phone fit and zoom the page
- Its inner pages carry a breadcrumb trail
- It declares an icon the browser can fetch
- It has a meta description of a usable length
- Its description says more than its title
- Its share card has a title, an image and an address
- Its share-card fields are valid and complete
- Its share cards are not truncated
- It gives X the fields a card needs
- It has at least 300 words of text
- It uses no obsolete HTML tags
- It has one main heading of a usable length
- Its headings run in order, with subheadings on a long page
- Its images use a modern format
- It links onward into its own site, not mostly off it
- It avoids inline styles, empty links and meta refresh
- Its HTML is not oversized in bytes or elements
- It declares its title, canonical and description once each
- It has a title of a usable length
- Its alt text describes the image
- Every link has text a reader can follow
- Each element id is used once
- It labels its form fields
- Its embedded frames say what is inside them
- Its images have alt text
- Its declared language is a tag a browser can read
- It declares its language
- AI answer engines are allowed to read it
- Technical 54of 106
- Presentation 21of 106
- Structure 18of 106
- Accessibility 9of 106
- AI answers 4of 106
The score is arithmetic you can redo
A page starts at 100 and each finding subtracts by severity. These four numbers are the whole model, drawn to scale, because a weight you cannot see is a weight you have to trust.
Three things we refuse to score
Every competitor prints a number here. We print the reason there is no number, because an invented one is worse than a gap.
- AuthorityNeeds a backlink index - referring domains, link quality. An audit reads your page, not the links other sites point at it, so any number here would be invented.
- Core Web VitalsNeeds lab or field performance data (LCP, CLS, INP). One HTML fetch cannot measure what a browser does after load.
- Content qualityWhether writing is good, deep, or matches intent is a judgement, not a measurement. We score the STRUCTURE of the content and say so.
- Technical SEOCrawls the live site and reports what is actually broken, with the URL attached.
- Keyword researchPulls real search data for your market rather than a paid tool's estimate.
- Structured dataReads the Schema.org markup on every page and writes what is missing. It will not invent a rating or an address to fill a gap.
- Internal linkingBuilds the actual link graph: orphan pages, click depth, and anchor text that says nothing.
- Answer-engine optimisationFeatured snippets, People Also Ask, voice answers and FAQ schema.
- Generative-engine optimisationGetting cited inside AI answers, engine by engine.
- Content optimiserProduces fully drafted pages for the gaps, not a list of topic ideas.
- Competitive intelPer-competitor analysis of what they cover that you do not. It needs the competitors named; in the recorded run the plan named none, so it produced nothing.
- AnalyticsAdvisory reporting. Read-only, and no account access is requested. A read-only Google Analytics link is built and switched off until Google approves our app; your Bing Webmaster Tools figures, read through Bing’s read-only sign-in, are built and available once switched on.
- OutreachDrafts who to approach and what to say. Nothing is sent. In the recorded run the plan named no one, so it produced nothing.
- Lead qualificationReads an enquiry you paste in and says how well it fits. It needs the enquiry, so the pipeline does not run it.
- Email deliverabilityChecks SPF, DKIM and DMARC before a single message is sent. Sending from an unauthenticated domain burns it permanently.
- Page speedMeasures response time, page weight and render-blocking assets. Where real Core Web Vitals cannot be captured it says so instead of guessing.
- ConversionWhat happens after the click: forms, calls to action, friction. No invented percentage lifts. We do not run your A/B tests.
- AccessibilityWCAG markup checks. Real AODA exposure for an Ontario business, and the same semantics search engines read.
- Local presenceThe plan for a business with a service area. The pipeline does not run it yet; the free local check in the app reads the same page.
- App store optimisationListing, keywords and screenshots, for mobile apps. It needs an app-store listing, not a website.
- YouTube auditChannel and video-level review. It needs your channel and a YouTube API key, not a website.
- StrategistPlans the campaign before a single doer runs.
- Build plannerTurns that strategy into the actual task graph, with dependencies.
- Quality reviewerScores every agent's output and recommends ship or rework.
- Campaign optimiserWould read results back and adjust the next cycle. It needs performance data from accounts you would have to connect, and we do not ask for them, so it does not run.
Most of the market means “Google and ChatGPT” when it says AI search. The generative-engine agent produces separate tactics for five, because they cite different sources and reward different things.
We publish that split rather than describing all 27 the same way. A firm whose pitch is “we tell you what we actually measured” does not get to be vague about its own product.
Your coding agent applies the fix. We still never touch your site.
Quinndatum speaks MCP. Point Claude Code, or any MCP client, at your account, and the audit stops being a document somebody has to read and retype. Your agent reads the findings, takes the patches, and puts them in your repository as a diff you review like any other.
We hold nothing, and that is the point
No account access. No write access. No script on your pages. Your agent runs on your machine, with your key, in your codebase. The change lands in your git history and stays there, including after you stop paying us.
That is the exact opposite of a tool that injects itself into your site and takes the improvements back when the subscription lapses.
It is told what not to invent
An agent handed only findings will confidently write all of them. Ours is handed the refusals too: what we would not score and why, which inputs were missing and whose limit that was, and the rule that a check whose input was missing did not run; it did not pass.
The patches arrive beside the changes we refuse to write (a title, a meta description, an H1) because each is a claim about what your business is for, not a fact about your markup. Your agent brings those to you instead of guessing.
No new price
It spends the credits your plan already includes, at the rates the app charges: two to read a page, three for its patches, so a complete cycle is five. Your plan’s monthly audit allowance applies too. On Free that is 5 complete cycles a month: the plan includes 200 credits and 5 audits a month, and the audits run out first.
A key reaches four routes and nothing else. It cannot see your credits, your plan or your account, cannot send anything, and cannot make another key. Revoke it and it stops working on the next request.