Find, fix, approve, then prove it
Five steps, each one shown
- 01
Visibility
One brand in. The prompts its buyers ask, and who the assistants name instead.
A brand starts with 10 prompts suggested from its own site — the best one for a need, one against another, what it costs, where — and you add, change or pause them. Each report, run when you start it or weekly or monthly on a schedule, asks them of ChatGPT, Claude and Google AI Mode, and stores every answer with the model that gave it, whether it named the brand and in what place, and the pages it cited.
What the engines answer, industry by industryVisibility · Acme Dental10 promptsSampleVisibility
17%
5 of 30 answers
Share of voice
10%
of names given
Average position
3.4
vs tracked rivals
Citations
3
answers cite the site
PromptEnginesChatGPTClaudeGoogle AI Modebest dentist for kids in AustinNamed instead: Bright Smile Dentalhow much does Invisalign cost in AustinNamed instead: Hillside Kids Dentaldentist open on Saturday in South AustinNamed on ChatGPT; elsewhere Juniper Family Dentistrydentist that takes Delta Dental in AustinNamed on ChatGPT and Claude - 02
Tasks
What each report says to do, one task per thing to change.
After a report the site is read for the prompts it lost: is there a page that answers them? Each gap becomes a task — a page to write or update, a listing to claim, reviews to ask for, a site file to fix — with the prompts it answers and the steps. A task stays until it is done; the next report adds to it rather than filing a second.
Tasks, for an agencyTasks · Acme DentalFrom the reportSampleWrite a pageSuggestedA page on what Invisalign costs, and what is included
Answers 3 prompts · 8 answers lost
Write it · 10 creditsUpdate a pageSuggestedSay on the kids page which ages you see
Answers 2 prompts · 6 answers lost
StepsListingSuggestedClaim the practice's Yelp page
Cited in 4 answers
- 03
Content
The pages written for you, and published when you say so.
An article is written from the site, its notes and what you add — never from facts it would have to invent; what it needs and cannot find is listed for you. Read it, edit it, or ask for a rewrite (2 credits). Approve it and a new page goes into WordPress or Webflow, with one click to take it back down; a section for an existing page, or a page anywhere else, you copy in. Autopilot writes a few a week if you switch it on; publishing them without review is a second switch, off by default.
What each platform lets you changeInvisalign cost in AustinIn reviewSampleHow much does Invisalign cost at Acme Dental?
Invisalign at Acme Dental costs from $3,800, set at one consultation, and covers every aligner, the check-ups between them and your first set of retainers.
What is included
- A 3D scan at the first visit
- Check-ups every six weeks
Left out
Not on the site or in your notes: how long treatment usually takes.
Publishes to WordPress - 04
Outreach
The pages AI cites that leave you out, and a way onto each.
Sources lists the third-party pages the answers cite, and whether each names the brand. Outreach takes the ones that leave it out: who to write to, from the page and the site's own contact page; an email drafted from the brand's facts, which you send from your own inbox; a listing to claim on a directory; a reply to draft for a thread. A page that names the brand goes Live, and the answers citing it show how they changed.
Outreach, step by stepCitation outreachTo doSampleMissing youNames Bright Smile Dental and Hillside Kids DentalThe 8 best family dentists in Austin
Cited in 6 answers to 4 of your prompts, by ChatGPT and Google AI Mode.
Email · 1 credit
To editor@austinfamilyguide.example, from its contact page
A practice for your family dentists list
Your list of family dentists in Austin is one of the pages assistants quote when parents ask. Acme Dental sees children from two, on Saturdays too…
Open in GmailCopy the text - 05
Proof
The same prompts, asked again, after the page is live.
We look for the page on the site until it shows up, and Proof notes whether Google had it that day — from Search Console when it is connected, otherwise from a Google search for its address. A change is asked again on its engine, up to four re-asks a Proof, and one that does not hold is not counted. The before and after goes into your client report, and you can run the whole report again yourself at any time.
ProofLive since 8 Nov 2026SampleInvisalign cost in Austin
Asked again 7 days after it went livePromptBeforeAfterhow much does Invisalign cost in Austin0 of 3WaitingInvisalign or braces for adults in Austin0 of 3Waitingbest Invisalign dentist in South Austin1 of 3WaitingNamed in 1 of 9 answers before it went live.
What this does not do
Each is something a tool in this category could be expected to have, and a decision not to build it.
Link building
A different business with different risks, and the reports do not show links deciding these answers. Outreach is narrower: getting named on the few pages AI already cites, by an email you send yourself.
Social posting
Engines do cite Reddit heavily. Posting into it on a client's behalf is not something we will automate: Community drafts a reply, and a person posts it.
Paid ads
Nothing here touches a budget or a bid.
Invented facts
An article says only what the site, its notes and you say. A price, a credential, a year or a result it cannot find is listed beside it, never made up.
Reusing a brand's own answers
A brand's own reports are always asked fresh, with the model that answered recorded. Only an identical prompt another brand asked in the past 7 days, on the same engine, may be reused.
Rewriting your existing copy
Updating a page means a section to add, never a rewrite of what is there. Your published voice is not ours to edit.
Why the loop is the product
Tools that measure AI visibility are a crowded category, and tools that generate pages a more crowded one. What almost nobody closes is the distance between them — and that is where the uncomfortable question lives: does adding a comparison page and some schema really make an engine change its answer? The last step is how that stops being a rhetorical question: the prompts the page was written for, asked again, beside the answers before it — and a change that does not hold when asked again is not counted, because an engine asked twice does not always answer the same. It is as capable of embarrassing us as of vindicating us.
Questions
- Where do the prompts come from?
- They are suggested per brand from what the business does, where it does it and who it competes with, then asked as written — and you can add, change or pause any of them. Language and location are inputs, not assumptions: a prompt about a plumber in Austin is not the same as one about a plumber in Manchester, and averaging them would hide the only thing worth knowing.
- Do you need access to the site?
- Not for the report. It reads public pages the way any visitor does, so there is no login, no plugin and no analytics connection involved in the diagnosis. Publishing new pages straight into WordPress or Webflow takes a credential you create there for the purpose — an application password, or the site's API token — and can revoke at any time; everything else is copied in, with no access of ours.
- How is this different from SEO?
- The output of SEO is a ranked list you can be sixth in. The output here is one answer, containing one to three names. The work overlaps — structure, claims, schema — but the target is a sentence an assistant composes, not a position on a page of results.
- Can you promise the engines will change their answer?
- No, and nobody can. What is promisable is the work and the measurement: the pages get published, the same prompts get asked again, and Proof says what happened either way — including when nothing moved.
- How long does the first report take?
- About 3 minutes. That is with the setup: you enter the site, check what we read from it and the prompts we suggest, and watch the report run. There is no queue to join and no call to book.
- When does Proof run, and what does it say?
- A week after a page goes live — found on the site, or published, where the site cannot be read. It asks the prompts the page was written for, up to 5, on the standard engines, and never reuses an older answer; then it sets each answer beside the one before the page went live. An answer that changed is asked twice more on its engine, and a third time if those two disagree; a change that does not hold is not counted. A Proof re-asks up to four times — new mentions before lost ones — at no extra credit, and marks any change past that as not re-checked. If the credits are short it says so on the article, and you can run it from there.
See who the engines recommend instead
The prompts your client loses, who wins them, and the tasks that would change it.