Including the bugs we shipped. A firm that sells diagnostics and hides its own defect history is selling something else.
This log starts on 29 August 2026. Quinndatum is new, and a changelog that pretended otherwise would be the first thing on it worth distrusting. It lists the notable changes, not every one, and it keeps the bugs we shipped. Entries are not rewritten later: where an old one is no longer true, a newer entry says so.
A new page lists every tool in the app: who can use it, its price in credits with the rule for when it is charged and what costs nothing, and what it runs on. Under the plans on the home page, the newest tools at their prices, with a link to it. Both are generated from the price table the app charges from, and checked against it before every deploy.
Free on every plan. Add a page you found and we read it once: whether it names your business (the name you give, or the one in your Brand facts), whether it already links to you, whether it is a resources page, and which of up to 10 of its outside links are dead. Canada’s anti-spam law puts the proof of consent on the sender, so for each contact address you give we keep a record: the page where it is published, the date we saw it, whether that page refuses unsolicited messages, and your note on why your message suits the role. We never collect addresses for you: you type each one, and we confirm it is on the page. On Growth and up, a short pitch drafted from your own page. Export it all as CSV and send it from your own mailbox: Quinndatum sends nothing. Three paid searches, for pages that name you without a link, pages that link to dead pages on sites you name, and resource pages on your topics, are built and priced, and switched off until we switch their paid data source on.
On Growth and up, free. Record a paid placement you arranged and we read the publisher’s page to confirm the link to your page is there, marked rel="sponsored" as Google asks of a paid link, on a page search engines may index; then we check it again every week, and the alert feed says when that changes. A link marked only nofollow fails, with its own reason: Google accepts it, but our rule is sponsored only. A placement buys the publisher’s readers and a mention of you, not rankings. Ordering a placement through us is not set up: we have not signed a placement vendor.
On top of your backlinks: the anchor text, which links are marked nofollow, sponsored or user-generated, the links gained and lost each month, and linking sites that share one IP range, from DataForSEO’s index and labelled as theirs. From the list you will be able to pick links for a disavow file, free, shown with Google’s warning that most sites never need one; Quinndatum never submits it. It is priced; it is switched off, with the backlinks themselves, until their prices are confirmed and that paid data source is switched on.
On every plan: the page’s links to other sites, sorted by a printed list of official and scholarly sites; its quotations, and whether they name who said them; its figures, and whether a source sits in the same passage. A 2024 study (GEO, KDD 2024) found adding these among the edits that most raised a page’s visibility in AI-written answers, on its own test engine and, for 200 of its questions, on Perplexity in 2023, not on any engine as it runs today; the page prints the study’s limits above the counts. It counts what is on your page and predicts no citation. A new AI topics tool, which sites AI answers cite for each of your topics from DataForSEO’s records, is built and switched off until we confirm its prices and switch that paid data source on.
Our page reader marked where a block of text starts but not where it ends, so in compact HTML the word after a closing block joined the next one: “One” ending one block and “Two” after it read as “OneTwo”. It touched headings in audits, the accessibility check, link anchor text and AI readiness. One malformed address on a page (a link, a canonical, a feed or a structured-data url) also failed the whole AI readiness, trust-signal or entity read; it is now skipped. If you watch a page, the list of changes after its next reading notes that our reader changed between the two readings, so a heading that now reads as two words is marked as possibly ours.
Uploading your Bing AI Performance export is its own tool under AEO and works on Growth and up without connecting Bing Webmaster Tools; below Growth its page explains why and keeps any exports you imported before. The Bing Webmaster Tools card is now only the Bing sign-in, marked not set up until that sign-in is switched on.
Paid data tools no longer say they run on our demo account. They say what is left before they switch on (confirmed prices, the data provider’s terms and a funded data account), that they will be on Growth and up, and that nothing can be sent or charged from your account until then. The free-first lines say that Search Console, Bing Webmaster Tools, your Business Profile and your Bing export are free from Google and Microsoft, and that reading them here is on Growth and up.
Bing’s AI Performance report shows how often Copilot and Bing’s AI answers cite your pages, and Bing offers no API for it, so you upload the file you export from it. The app shows which of your pages are cited, how often, and the grounding queries behind them, with the dates the file covers and the day you exported it. These are Bing’s counts, labelled as Bing’s; Bing calls the grounding queries a sample, and so do we. Free, on Growth and up. A newer file for the same dates replaces the older one, and you can delete any file. The site’s AI visibility overview shows them too, beside your Search Console traffic once that is connected, each under its own source and dates.
Before a paid data tool, the app now names the free source that already measures your own site: Search Console, Bing Webmaster Tools, your Bing AI Performance export or your Business Profile, linked where it is set up. Those sources describe your own site only; where a paid tool compares you with competitors, the line says the free source does not show them.
Bing Webmaster Tools: your clicks, impressions and Bing’s own position for your searches and pages, the links Bing has found to your site, and its crawl problems, read through Bing’s read-only sign-in, which you allow on Bing’s own page; we never ask for an API key. It waits on our registration with Bing. Your Business Profile: your own Google reviews and your replies, read on each visit and never stored, and never answered for you; Google offers no read-only access, so the app says that too. A Google rating lookup for any business, competitors included: its rating, review count and the up to five reviews Google returns, with each reviewer credited as Google requires, at the audit-page price and within a monthly limit that keeps it inside Google’s free allowance. The Business Profile link and the rating lookup wait on Google. Each is refused, with nothing charged, until it is switched on.
Which pages Google’s AI Overview cites for each of your phrases is now read from Google’s results page itself, through DataForSEO, rather than from DataForSEO’s database of AI answers, which carries a monthly minimum. When Google loads the overview after the page and its sources are not captured, the app says it does not know, never that you are not cited. It is built and switched off until that paid data source is connected; ChatGPT and Perplexity stay on the database route, also switched off.
The trust-signal check lists what a page shows a reader about who stands behind it: HTTPS, who publishes the site, a named author and dates on articles, about, contact, privacy and terms pages, credentials stated in its structured data, links out to sources, and review markup, including whether Google treats it as self-serving. It is a list with where each was found, never a score: Google says E-E-A-T is not itself a ranking factor. The entity check reads the organisation your home page declares, whether your About page agrees, and whether the profiles it lists name you back; a profile that shows our reader a sign-in page is “could not read”, not a mismatch. Each costs what an audit page costs.
On every plan, at the audit-page price: the share tags a page carries and the card each of Facebook, LinkedIn, X, WhatsApp and Slack builds from them, with the image fetched and measured from the file against each network’s own published rules, each linked. On Growth and up, posts for each network drafted from a page, checked against your own site and brand facts before they are shown; a sentence with a figure your page and brand facts do not state is removed. We never connect to a social account: you post them, or not.
Quinn’s fixes now carry the steps for the platform the page was built with, read from the page itself: WordPress (with Yoast or Rank Math where we see them), Shopify, Wix, Webflow, Squarespace, or plain HTML when we cannot tell. The steps come from each platform’s own help pages; where we could not confirm a menu name, the step says to check it in your editor. They are in the app and in the downloaded report, inside the price of the fixes. We change nothing on your site.
A site audit now reads up to your plan’s page limit in one run: 5 pages on Free, 50 on Starter, 250 on Growth, 500 on Pro and 1,000 on Agency, a few pages at a time and each started at least half a second after the last, so your server is not rushed. The page table adds each page’s title, description and first heading, word count, links out, response time and size, and a CSV file with one row per page. A watched page can be re-read daily as well as weekly, and the app lists what changed: its status, whether it can be indexed, its canonical, its title or its score. A reading the site did not answer at all is our failed read, not an alert.
One page per site with what we measured that bears on AI answers: which AI crawlers its robots.txt lets in, its llms.txt, how much of a page is there before any script runs, and its structured data, each with the date, the page and how it was measured. It costs nothing, reads only what your audits and AI readiness checks already stored, and has no combined score. Its paid parts stay switched off, as below.
Your own Google Analytics figures (sessions, engaged sessions and key events by landing page and channel, beside your Search Console figures and able to fill the return-on-investment calculator), through a read-only link you would grant on Google’s own page, wait on Google’s approval of our app. A lookup of your brand in Google’s Knowledge Graph waits on our key for it. From DataForSEO: keyword suggestions, a competitor keyword gap, the AI Overview beside each rank reading, who holds the answer box for each phrase you track, and a local map grid of where your listing appears in Google Maps around your business. All are refused, with nothing charged, until connected; none is on sale today.
On the paid backlinks tool (switched on for our own demonstration account only), the 62 came from DataForSEO’s list of referring domains and the 61 from its separate summary, so the sentence read more than the whole. A count read is now only ever out of the list it came from, and the summary’s own count is named as the summary’s.
One page, read as it is sent, before any script runs: whether its text is there at all, whether the first sentence under its title and each question opens directly, whether its sections stand on their own as passages, its forms and links, its feeds, its AMP version, and which AI crawlers its robots.txt lets in. It costs what an audit page costs and runs no AI model. It also gives you plain “summarise this page” links for five assistants, with no hidden instruction aimed at the reader’s assistant. For AMP it reports whether the version loads and points back; it does not claim a ranking benefit.
On every plan, at the audit-page price: which ad and analytics tags the page’s HTML loads, whether a click id and UTM tags survive its redirects, how many words of your ad the page repeats, and how far down the first form, phone link or button sits. It never touches your ad account. The calculator is free and works only from the numbers you type, with every formula shown; it is arithmetic, not a forecast.
The words and phrases up to three competitor pages use that your page never does, counted from the pages themselves, at the audit-page price. Other tools sell this as “LSI keywords”; that name claims a ranking mechanism we cannot show, so ours says what it is: a count, not a ranking factor.
Which phrases from a demand gap a page already answers in other words; what a page you name says that no passage of yours comes close to; the people, organisations, products and places a page names; and Ask your site, which answers a question only from your own pages and quotes the passages it used. The fan-out check has a language model list the narrower questions an AI search could split one question into, labelled as the model’s list, and shows which of your passages match each. The app listing check reads a public App Store or Google Play page against each store’s limits and suggests rewrites, with any figure the listing does not state removed. AI-written text is checked against your own pages before it is shown, as the AI specialists’ drafts are.
Type your business name, what you offer, your prices, the certifications you hold and those you do not, and where you serve. The AI specialists are given them as true, and a draft that claims a certification you listed as not held is blocked.
Which AI answers cite or name you and your competitors, your share of them, the answers that cite a competitor and not you, which of your search phrases show an answer box, and a reader for your Google reviews: all from a paid data source (DataForSEO), and all refused, with nothing charged, until that source is connected. Your own Search Console figures, through a read-only link you would grant on Google’s own page, wait on Google’s approval of our app. None of these is on sale today. Citation monitoring through logged-in assistant sessions stays not offered.
Until now the home page, the sample audit and the terms gave four different versions. There is one now. A one-off audit is refunded in full only if it finds nothing you can act on; tell us within 30 days of delivery. Monthly plans, including Watch, cancel any time, with no partial refunds.
An example on the home page named a real company’s site, without its permission, with figures that were not true of it. It is gone. Only sites we own are shown now, and a check fails the build if any other business is named in our copy.
Citation monitoring was marked as working. It has never taken a reading and is not offered. Seven auditors were marked as not shipped although they produced output in a live run on 4 October. Each agent’s mark now comes from that recorded run, and the legend counts the marks instead of being typed.
The home page described launching ad campaigns, posting to social accounts, publishing to your site and sending cold email. None of them can be started, and each would need logins we never ask for. They are now marked as not offered.
Free includes five audits a month, and that runs out before the credits do. The MCP page and the home page now say so, and a check holds both numbers to the plan.
Every price on the site is in US dollars. The structured data said Canadian dollars; it now agrees with the page.
Free and Starter run the site audit and its deterministic checks. A model run is refused on those plans before anything is read or charged, and the message names the plan that unlocks it.
A site’s headline score is the average of its indexable pages, with how many pages it covers stated beside it. The fixes queue charges only for a patch it actually wrote.
Each area (SEO, AEO, Content, Local, Research, Outreach) has its own overview and menu, each site has an overview of its own, and every report can be shared by private link or exported as Markdown, CSV or PDF. Five more tools run: keyword ideas from your own seed terms, sizing up competitors you name, grading a draft you paste, drafting outreach pitches (drafts only; nothing is sent) and reading an enquiry you paste.
Seventeen specialists run on one site in order, each reading the deliverables before it: a plan, a work order, eight specialists, six auditors and a reviewer. You are charged 5 credits as each one delivers, never for one that fails. The four that would send or spend for you never run here.
One light style across every page and screen, with a dark theme in the app.
Who runs this, where, what we build and what we refuse to build. Read it.
Opening an audit that had not paid for Quinn’s fixes wrote them and charged 3 credits. Now a button names the price, nothing is spent until you press it, and reopening the audit afterwards is free.
Make a key in Account and hand it to Claude Code or any MCP client. A key reaches four routes and nothing else, spends your plan’s credits at the app’s prices, and stops working on the next request after you revoke it. Setup.
Keys were left behind when an account was deleted. No key was left working without an account behind it; the leftover records were removed, and the deletion receipt now lists keys with everything else.
A page we could not read now costs nothing and uses none of your monthly allowance, and the report says which pages we could not read and why.
When a check’s input was missing, it was shown as passed. It now says it did not run.
Two copies of the background watcher could pick up the same page. A reading is now claimed before it starts, so only one can take it.
The audit grew from 29 checks to 49, each tested on a page that should trip it and one that should not. Four fired falsely on real sites and were fixed before release.
A one-page score can say what is wrong with a page. It cannot answer which page do I fix first, which is the question a site owner actually has. The site audit takes pages from your sitemap, scores each with the same checks and the same published weights, and ranks them. It is capped by whichever runs out first — our web limit, your credit balance, or your plan’s monthly allowance — and the result says which. The pages it did not read are reported as unread, not as passing.
Heading order, duplicate tags, X cards, hreflang validity, URL hygiene, page
weight, image format, markup hygiene, and whether anything is measuring the page.
Plus a check for robots.txt blocking GPTBot,
ClaudeBot, PerplexityBot and nine others — if you are
blocking the crawlers that feed AI answers, nothing you publish can be cited in one.
Every new check is tested twice: once on a page that should trip it, once on a page
that should not, because a check that cannot fire is a number on a sales page.
Whether your four address forms redirect to one, whether HTTPS answers, what your sitemaps contain, and which AI crawlers are blocked. Reporting “www and non-www both answer” on every page would be a hundred copies of one problem.
The duplicate-title check counted <title> elements inside
inline <svg> — which are the accessibility labels that make a
chart readable to a screen reader, not document titles. On a real site it reported
three titles where there was one, and the fix it recommended would have
removed an accessibility feature. Caught before release by checking a surprising
finding against the live page instead of believing it.
The coverage line said 10,806 pages and the domain panel said 11,125, because two different pieces of code walked the sitemap separately. We have spent this build criticising a competitor for reporting four different page counts on four screens. Ours were on one screen. There is now a single traversal, and the test asserts the two numbers are equal rather than merely present.
It was eleven screens tall on a new account — the playbook list alone took four of them — and the left rail navigated by scrolling. The rail now switches the screen: one screenful when you arrive, four and a half for the longest. Nothing is hidden behind a script, either: if the page fails to load its code, every section is still reachable by scrolling, exactly as before.
The check looks for rules blocking GPTBot, ClaudeBot and
the rest. When there was no robots.txt at all it found no rules, and a
check with no findings was rendered as one that passed — a conclusion
drawn from a file we never opened, on the single question this product is most about.
A check whose input is missing was not evaluated, and it now says so in those words.
It was near-black while every other page here is light, so signing in crossed a theme boundary for no reason. Measured after the change: 466 text readings across every screen, none of them below the contrast minimum. The old dark palette is kept intact rather than deleted.
The navigation rail measured 533px inside a 390px viewport, so the whole page could be dragged off-centre. Nothing looked broken, because the rail scrolls — and our test for horizontal scrolling only ever ran at desktop width. It now runs at phone width too, which is where it failed.
It counted every sitemap URL its crawl had not reached as an orphan. With a 25-page cap against a 404-page sitemap, that is 353 false findings — and it failed in the direction that manufactures alarming results, the kind a client pays to fix and then discovers were never broken. The engine now refuses the number and says why. The refusal is shown in the sample audit.
It truncated a domain's TXT records to twenty before filtering for
v=spf1. Against a domain publishing 28 records, the SPF entry fell outside
the slice on roughly a quarter of lookups — and the agent then declared the domain
unsafe to send from. Found by an adversarial check, not by a customer.
The runner kept a hand-maintained list that had fallen behind the registry, so it
said 17 pass, 0 fail without touching a quarter of the system. It now
refuses to run at all if any registered agent is missing from it.
Structured data, internal linking, page speed, conversion, email deliverability and accessibility. The last two cover ground no content-publishing tool audits, because neither is fixable by publishing a page.
We ran the engine on our own site and posted everything it found, including a domain that could not receive email. Read it.
Audit, audit with a 30-day re-check, and a monthly Watch. Refund window went from 7 days to 30, and if every finding is one you already knew, it is free. Replaced on 10 October 2026 by one refund rule; see that day.
Marked agent by agent on the home page. Eleven have produced real output; the rest are built and connected and have not yet been used on a paying engagement. Corrected on 10 October 2026: there were no paying engagements, and the marks were wrong both ways; see that day.
We do not do the work, do not monitor continuously, do not promise rankings, and never ask for account access. (Of the $149 audit. Watch, and the self-serve app’s watched pages, do repeat.) The accessibility check is markup only and the speed figures are lab measurements — both labelled as such in the report.