The report that shows its work.
Type a domain and the free check runs now: up to 25 pages, 5 queries against live Google results, the link summary, and every check it ran with its value and its result. No account, no email, no card.
The paid run goes further. It connects to your Search Console with read-only access, crawls the site, and writes a document in which every sentence points at a measurement you can open. 149 dollars. No call, no onboarding, and nothing to sit through.
What the free check does, and what it will not do.
It crawls up to 25 pages, starting from your sitemap and your homepage. It checks 5 queries against live Google results. It reads the link summary: referring domains, backlinks, spam score. It reads your robots.txt, your sitemap, your homepage, and the noindex and canonical directives on the pages it fetched. For each of the 5 queries it tells you whether your domain appeared in the AI Overview’s citation list. Then it shows you every check it ran, the value it measured, and whether it passed.
It does not connect to your Search Console. You have not given it access to anything, so it cannot, and every number it prints is a third-party estimate or a single observation. It labels each one as what it is. It will not print an answer-engine score, because one observation of an answer engine is not a measurement. It does not crawl your whole site, it does not keep history, and it is not a preview of the paid document with the good parts removed: it is a different instrument measuring a different thing, and section 3 explains the difference.
Free tools see a fraction of your site. You can watch it happen on yours.
Every third-party database, ours included, indexes the queries its own crawlers track at a volume worth tracking. That is not a defect in the tool, it is the definition of one. A site whose traffic arrives from thousands of small queries is nearly invisible inside that index, and inside it the site looks the same whether it is healthy or half missing from Google.
We know because we made the mistake ourselves. We concluded from a third-party count that most of one of our own sites was not in Google’s index, and we published that conclusion. Google’s own reporting for the same property, over the same weeks, showed an order of magnitude more pages and more queries than the third-party count had. The third-party tool had answered a slightly different question from the one we asked it, and we had not noticed. We corrected the published figure in place, with the date, and the correction is still on the site.
That account is anonymized at the owner’s request, which means you cannot check it, and you should read it as something we are telling you rather than something you have seen. What you can check is your own site.
Run the free check on the domain you are considering. It shows you every check it ran, the value it measured, whether it passed, and which of its own numbers are third-party estimates rather than observations. It also prints the denominator behind every aggregate: how many pages its index reached, against how many your sitemap lists. If your traffic arrives from thousands of small queries, that gap is this section’s whole argument, on your own domain, in about a minute.
One thing it cannot do, and this is the boundary you are being asked to pay to cross. The free check has no access to your Search Console, so it cannot show you Google’s own numbers for your property, and it never will. The paid run asks Google first and treats third-party data as an estimate of it.
What arrives for 149 dollars.
One document, at a private URL that only you can open, plus a PDF of the same content for anyone you have to convince internally. It has five sections and an appendix.
- 1. The verdict.
- One paragraph naming the constraint the run identified: indexation, click-through rate, authority, or topic selection. If your site is too new for a diagnosis to mean anything, it says so, gives the date at which one becomes meaningful, and states the reason.
- 2. The findings, ordered by strength of evidence.
- Contradictions first, then indexation and access, then your own search performance, then authority and internal links, then whether the queries you care about are winnable, then your visibility in AI answers, then public files and store listings, then Core Web Vitals, then what the run cost. The order is not a style choice: a click-through-rate finding on a site with a broken indexation measurement is not a finding, it is a consequence, so a finding whose prerequisite failed is not printed at all.
- 3. What the data does not show.
- Generated from the run’s own record rather than written: the checks that could not run and the reason for each, the sources whose coverage fell below the threshold, and the facts no sentence used. It cannot be empty, and it cannot be shortened quietly, because nothing in it is composed by the writing model.
- 4. What this analysis cannot tell you.
- The refusals, printed only where they bind on your run.
- 5. What the investigation cost.
- The sum of the cost field the data provider returns with every response, the number of calls by endpoint family, the wall clock, the number of facts, and the number of checks.
- The appendix.
- Every fact of the run, including the ones no sentence used, with the provider’s request identifier and the time the provider answered. It is what makes the rest of the document checkable.
Every sentence points at a measurement. The ones that did not were deleted.
Every fact the run collects is stored with the check that produced it, the provider’s request identifier, and the time the provider answered. Each fact gets an identifier derived from its own contents plus a random value generated at the start of that run. The writing model is never given that random value, so it cannot compute an identifier, and the only valid identifiers it can produce are ones copied from the list it was handed. An identifier it invents is rejected.
The document contains two kinds of sentence, and they are printed in visibly different styles. A claim sentence is generated from the check’s own template with the fact’s values dropped into it, so it says what was measured and nothing more. An interpretation sentence is written, and it is kept only if every number in it appears in a fact it cites, every name in it appears in a fact it cites, it cites at least one and no more than eight facts, and it contains no causal verb at all, because causation is exactly what a measurement cannot establish. Sentences that fail are deleted. If a whole paragraph fails, the paragraph disappears, and the fact it was carrying is printed anyway in the check’s own words, because those sentences are generated from data rather than written.
Two counts are printed with the document: how many checks ran and how many could not, and how many sentences the gate deleted.
One more thing, because it is the part we would rather you did not have to find out for yourself. The gate guarantees attribution, not truth. A sentence that cites a real fact and uses approved vocabulary can still mislead by what it leaves out. So the document also prints the list of findings the interpretation did not address, and the appendix prints every fact of the run, including the ones no sentence used.
What it will not do, and what that costs you.
It will not tell you which topics to write about. Choosing queries needs three judgements a threshold cannot make: whether the question can be answered in a single sentence, whether it carries buying intent, and whether it is relevant to your business at all. Relevance is where it fails, and it fails loudly. Expanding a keyword list one step too far once put three unrelated queries at the top of the list, two of them with monthly volumes in the hundreds of thousands, for a product they had nothing to do with. The highest-volume one would have passed every difficulty filter we apply, because the first page was not dominated by one player. It was simply about something else. So the run measures the queries you give it, plus a default set drawn from queries you already rank for, and it names this as the largest gap in what you are buying.
It will not tell you whether your content is good, or which pages to delete. No data source we use measures whether a page is useful, original, or written for a person rather than a crawler, and a page with no position in Google can still be the page an AI answer quotes. So the run reports whether you were cited when Google answered the question itself, and it reports what your pages earned in clicks, and it leaves the decision about the page to you, because deleting a page is the one act here that cannot be undone.
It will not write your replacement title. It will tell you that the title is not answering the query that reaches the page, which query that is, and the measured character budget for that page type on your site. The budget is measured and never copied out of your code, because the two are routinely different: Google appends a suffix of its own on some sites and then truncates the result, and that suffix appears nowhere in your source. A budget read from a constant in the code is a budget that is too long.
It will not judge whether a link is worth acquiring. It gives you the gap list with the spam filter applied. Our own expectation, stated in advance: on small products, about ten of two hundred domains survive the filter.
It will not rule a manual action or a security issue in or out. Those two reports are interface-only, and the Search Console API we use does not expose them. What you get instead is the shape of the trend, because a sudden large fall and a gradual one are different findings, plus the crawl’s status codes. That narrows it and does not settle it.
It will not tell you that a fix worked. It will tell you the effect is not visible in this data, which is a statement about the measurement and not about the world.
It will not review your design, or whether a page looks trustworthy.
And if you do not connect Search Console, the entire first-party section does not run, the document falls back to third-party substitutes, and it labels every one of them as an estimate rather than quietly printing them as facts.
There is no sample. There are two things you can run.
There are no customers yet, so there is no customer document to show you, and we are not going to publish an anonymous one. A document about a site you cannot look at would ask you to take our word for the one thing this product exists to stop anyone taking our word for.
Two things on this page are real, and both of them are free.
The free check runs on your domain. It shows you every check it ran, the value and the result, the denominator behind every aggregate, and which of its numbers are estimates. It is built from the same check catalogue as the paid document.
The claim checker takes a report somebody sent you, pasted or uploaded, and sorts every claim in it into three groups: the ones public data supports, the ones nobody outside your business can check, and the ones public data already contradicts. Every claim comes with a receipt you can open. It runs the same gate the paid document runs on itself, so what you are watching is the product, pointed at a different document.
What you cannot do is read a finished document before you buy one. The first finished document will belong to the first customer, and it will be theirs to publish or not, not ours. If two free instruments are not enough to go on, they are the whole of the pre-sale and you should stop there.
149 US dollars, or 349 with an answer-engine measurement that can be repeated.
149 US dollars. One document, one property, one run. Whole-site crawl, the Search Console checks, the search performance analysis, the interpretation, the fact table, and the statement of what the run cost. Nothing renews and there is no plan to cancel.
349 US dollars. The same document, plus a larger crawl with page-level parsing of your key pages, and a replicated answer-engine measurement: 25 prompts, each run 8 times. It covers Google’s AI Overviews only. It does not cover ChatGPT, Claude, Perplexity, or the Gemini app, and the document says so next to the number rather than in a footnote.
The reason the second tier exists is that an answer-engine figure from a single observation is not a measurement. Independent work published in 2026 put the number of answers needed before a ranking stabilized at between 33 and 94, with a typical margin of error of about five positions for a top-ten site and one case in five above ten. The trade body’s own guidance classifies fewer than 50 queries as exploratory, below the level at which a category can be characterized at all. So every answer-engine figure in the 349-dollar document is printed with its repetition count, and the document will not report a difference smaller than the stated margin of error.
Why the numbers are in dollars. The work is billed in US dollars, so that is the currency the prices are written in. Stripe converts at checkout and charges your card in your own country’s currency. The rate you are shown includes a conversion fee of between two and four per cent, which Stripe sets and keeps; it is the reason the figure on your receipt is not the figure above. Your bank may add a foreign transaction fee of its own, and we cannot see it or predict it.
Tax is added, not included. Whatever your country charges on a sale like this is calculated at checkout and added on top of the price. It is not part of the figures above. We are not telling you what it will be, because the rate depends on where you are and, in some countries, on what kind of buyer you are. Stripe calculates it, collects it, remits it to the tax authority, and shows it as its own line before you pay.
What the run costs us. 83 US cents of data and language model for the 149-dollar document. About 1.08 US dollars for the 349-dollar one. Those are in US dollars because the data provider bills in US dollars. They are not converted for you and they are not part of what you pay; they are printed because the two figures normally hidden from a buyer are the cost of the work and the limits of it, and this product prints both. The exact figure is in your document, summed from the cost field the data provider returns with every response, so you can check it against the provider’s published price list. We do not print a notional cost for our own time, because that number would be unverifiable and it would become a floor under every future price.
Paying. Checkout is Stripe, and the tax, the receipt and the invoice are handled there rather than here. You can pay as a guest. The receipt comes from Link and the charge appears on your statement as LINK.COM, so that is what you will see, not our name.
Where this can be sold. We take payment from most countries. Stripe acts as the merchant of record and handles the sales tax, VAT or GST on the sale in more than eighty of them. There are countries where it does not, and in those the tax on the sale is our responsibility rather than Stripe’s. The list is in the Terms, and it is worth thirty seconds of your time if you are in a country whose tax rules you already know are unusual.
The second run. Ninety days later, the same checks run again on the same property, and the document reports what changed, check by check, and whether the effect of each recommendation arrived inside the window the first document named. Where a window has not closed yet, it says so rather than reporting a null. It is the same product bought again, at the same price: 149 US dollars, or 349 US dollars if you want the replicated answer-engine measurement in the second document as well. There is no subscription, no renewal, no discounted re-run price, and nothing to cancel. You buy it in ninety days, or you do not.
Who should not buy this.
You want the work done. There is no implementation, no retainer, no outreach, no content writing, no link building, and no one to hand the document to. It is a document.
You want a guarantee. Nobody can give you one, and the search engine says so itself: “If they guarantee you that their changes will give you first place in search results, find someone else.”
You want a verdict on your agency. This measures your site. It does not assess another company’s work, and it will not name one.
Your site is too new. If Google has no impressions to report for you yet, the verdict will say so, name the date at which a diagnosis becomes meaningful, and give the reason. First impressions typically appear two to six weeks after indexing and a stable position takes three to six months. You would be paying for a document whose main finding is the date. Run the free check first and come back.
You want to talk to someone first. There is no call, before or after. The free check and the claim checker are the whole of the pre-sale, and the document is the whole of the delivery.
Your Search Console access, in one paragraph.
The paid run asks for read-only access to your Search Console property. Read-only means it can read and cannot change anything. You grant it through Google, you can revoke it from your Google account at any time, and revoking it mid-run is a state the product handles rather than a state that breaks it.
Your numbers appear in your document and nowhere else. There is no comparison of your property against another customer’s, no sector benchmark, no aggregate, and no “see how you compare” feature, because the terms we operate under forbid turning your data into a dataset, and because a benchmark would require us to hold data that is not ours. The document lives at a private URL. There is no public version of it and no shareable link that includes your first-party figures.
We keep the raw provider responses and the fact table so that any sentence in the document can be traced back to what the provider returned, for as long as the document exists. Ask us to delete everything and we will.