In private build
Six case studies on properties I own produced one conclusion I could not buy a tool for. This is what I am building instead, published as a bet with a date on it rather than as a launch.
It connects to Google Search Console, reads what the product actually does from its own documentation, changelog and pricing page, and sends five prioritised actions a week ranked by trials and pipeline rather than by audit severity. It is in private build. There is no name yet, no public pricing and no launch date, so this page is the thesis rather than a product page, and the section at the bottom lists what would prove the thesis wrong.
How it ranks a week of work
search opportunity × commercial intent × product relevance × conversion potential × ranking feasibility ÷ implementation effort
Six terms, and the divisor is the one most scoring models leave out. A fix worth a lot that needs a sprint from an engineer who does not have one loses to a fix worth less that a founder can ship tonight. Product relevance is the term that needs the knowledge layer. Without it a keyword with volume and intent still scores high on a query the product cannot serve, which is how a SaaS blog ends up ranking for work it cannot close.
Opportunity and hypothesis
Every case study in this section publishes the bet it made before the data existed, and the part of that bet which failed. This page does the same thing in the only order available to it: the bet first, because the data does not exist yet.
Every SEO tool I pay for is built for a site that sells attention. It grades pages against rules and sorts the work by how badly a rule is broken. Software does not sell attention, it sells a subscription, and the queries that end in a subscription are product comparisons, alternatives, integration questions and pricing. None of the tools knew what my product was, so none of them could tell me which of those queries I could actually serve.
That a persistent product knowledge layer, built from the docs, the changelog and the pricing page, changes what a prioritised list is worth. Severity is a property of the page and any crawler can compute it. Commercial value is a property of the query, the product and the buyer together, and it cannot be computed without knowing all three.
Nowhere, and that is the problem. The thesis has not survived contact with an export because there is no export yet. Every other page in this section earned the right to make a claim by publishing the data behind it, and this one has not. Treat the argument as a stated bet with a date on it, not as a finding, until the numbers exist.
A founder with no marketing team gets a week of work they can actually ship, and the reporting ends at MRR instead of at impressions. If it does not hold, the failure will be visible in the same place the successes would be: the actions get shipped and the trials do not move. That is the number worth watching and it is the one I will publish.
The build
A feature list is easy to write and impossible to check. Each part below names what it does and then what it exists because of, and every figure in the second column is one this domain publishes somewhere else.
Five things to do this week, in order, in plain language. Quick wins, missing pages, cannibalisation, internal links and conversion fixes all compete for the same five slots on the same score, so they are finally comparable instead of sitting in five separate tools.
Why it exists
A 400-row issue export is not a plan. I have opened plenty of them twice and never a third time, and the rows that got fixed were the ones that were easy rather than the ones that were worth it.
Runs the site against three competitors and lists the commercial page types that are missing: alternatives, comparisons, feature pages, use cases, industries, integrations, migrations and templates.
Why it exists
On the software property in this section, 52% of the 241,500 Copilot citations landed on commercial-intent queries rather than informational ones. The pages that answer those are the pages a content calendar built around topics never reaches.
Crawls the marketing site, the blog, the documentation and the app routes as one graph rather than four, queues named links pointing at conversion pages, and surfaces the orphans.
Why it exists
On 2026-08-21 I read my own link graph and found 118 of 122 money pages with no link from any post, and 12 of 13 posts not linking to their own hub. A day later an audit of this very section found 55 internal links arriving at the six case studies and not one leaving for a commercial page. I did not need a crawler to tell me that. I needed something that queued the link.
Maps every Search Console query cluster onto a feature, an integration, a use case, a doc or a landing page, and returns the mismatches: demand with no page, and pages with no demand.
Why it exists
That mismatch list is what tells a founder whether the positioning or the content is the thing that is wrong. It is also the only report that can catch a keyword with real volume and real intent sitting on a query the product cannot serve.
You paste in real answers from ChatGPT, Gemini, Perplexity, Claude or an AI Overview and it pulls out the cited domains, which competitors got recommended, the claims your pages do not make, and the pages that would have to exist to be cited at all.
Why it exists
Pasted answers are a deliberate design choice, not a shortcut. None of those four engines publishes a per-site citation report, which is the same limit every study in this section states in its own scope note. A product promising continuous monitoring of them is sampling a scrape and charging for it.
Turns the way people actually complain about the problem into problem-aware queries, comparison searches and landing page copy.
Why it exists
Keyword tools return the vocabulary of the category. Threads return the vocabulary of the buyer, and for a product in a category that did not exist eighteen months ago those are not the same words. The category vocabulary has not been written yet.
What it reads
What it reports back
One chain, end to end: query, landing page, signup, activation, subscription, MRR. Split brand against non-brand, and self-serve against sales-assisted, because a founder deciding whether search is working needs those four numbers separately and an impressions chart gives them none of it.
It also watches the two failure modes that kill SaaS content quietly: decay on pages that used to convert, and product drift, where the page still describes a version of the software that shipped eleven releases ago.
Falsifiable
Published now, before there is anything to report, so the first export can be checked against them rather than interpreted next to them. If the right-hand column happens, the thesis was wrong and this page stays up saying so.
Stated 2026-08-23. There is no data behind any row yet, which is the point of stating them.
| The prediction | How it gets measured | What would falsify it |
|---|---|---|
| Docs hold more internal authority on a SaaS site than the blog does | Internal links and URL Rating distribution across /docs/ versus /blog/, per property | The blog outranks docs on internal links and the docs turn out to be a dead end |
| Commercial page types beat topic posts on trials per session | Trials attributed to alternatives, comparison and integration pages versus blog posts | Topic posts convert at the same rate or better once volume is controlled for |
| Citation share on answer engines moves before organic clicks do | Bing AI Performance citations against Bing organic clicks, same pages, same window | Clicks and citations move together, which would make the extra measurement pointless |
| Five scored actions a week outperform a full audit backlog | Actions shipped per week, and trials 90 days on, against the same team before the tool | Teams ship the same volume either way, meaning the ranking added nothing |
| A product knowledge layer changes which keywords get chosen | Overlap between what it picks and a volume-and-difficulty pick on the same site | The two lists agree, in which case the knowledge layer is expensive decoration |
The honest part
Building a tool around a method does not exempt the tool from the standard the method was published under. That standard is the reason this section holds six case studies rather than twenty, and the reason this page is not counted as one of them.
Questions
Keep up with the tracking
The numbers get published as they come in rather than once they are tidy. The studies in this section are refreshed against new exports, the citation figures get re-pulled, and this page gets a result column when there is one. That is the part worth following, and it is also how you will find out what it ends up being called.
Sent when there is a number worth sending. Early access invites go to that list first.
241,500 Copilot citations against roughly 2,500 organic clicks, with both dashboards on the page. Worth reading before deciding whether the premise holds.
1.42 million impressions, 8,666 clicks, a 0.61% click rate, and the citation figure next to it. The failure that made the scoring model necessary.
The same method, run by hand. It is what the tool is being built to automate, and it is available now.
Back to the hub, and the evidence everything above is derived from.