A playbook in action
When shoppers ask AI what to buy — be the answer
Four playbooks, one billable loop. Claude drafts fifty real shopper questions from the brand kit, you approve and lock them, and they run live across ChatGPT, Perplexity, Gemini and Claude. You get the shelf, the gaps and the reputation read — then the corrected llms.txt written against exactly what you’re losing, and a white-label report to forward. Re-run monthly; the deltas are the evidence.
New here? Connecting Claude takes about a minute →The loop, in four moves
400 tokens a run, and Claude confirms the price before it charges.
Step 1 · You approve the questions
Fifty questions your client’s buyers actually type
Claude drafts the pack from what the brand kit already knows — five lanes, real shopper phrasing — and walks it past you lane by lane. You edit, you name the competitors, you sign off. Then it locks, because a measuring stick that changes shape measures nothing.
The whole ask — one message
“Draft the 50-question AI visibility pack for Acme and show it to me lane by lane before saving.”
How the pack is built
Fifty questions the way real shoppers type them — buying intent, brand, comparisons, use cases, reputation — drawn from what the kit already knows about this client.
Claude shows the pack lane by lane and takes your edits. Nothing runs until you've signed off the questions and the competitor list.
The saved pack becomes the measuring stick — every re-run asks the same questions, so month-over-month movement is a fair comparison rather than a reshuffle.
Step 2 · The engines answer
Two hundred live answers, and the shelf they put your client on
Every question runs across all four engines with web search on — the answering mode shoppers actually see. What comes back is mention rates per engine and lane, the share-of-voice shelf against the competitors your client named, and the questions they’re losing outright. From the second run on, every number carries its movement.
Four engines, asked live
Then just ask it what happened
“Where does Acme show up, and on which engines?”
“Who do the engines recommend instead of us?”
“What do they say when a shopper asks if we're legit?”
“Which questions are we losing outright?”
Step 3 · You ship the fix
A corrected llms.txt, written against the gaps you just measured
This is the part a rank tracker doesn’t have. Claude fetches the client’s existing file and treats it as the incumbent — improving it, never replacing it — then writes the new layers against the exact questions the pulse showed they’re losing, grounded in their live site and brand kit.
One ask, the whole fix
“Write Acme's corrected llms.txt against the gaps — use the existing file at acme.com/llms.txt as the base.”
Receipts, so the work is visible
The client sees exactly what changed without diffing files — and the next pulse keeps score.
Step 4 · You bill for it
The white-label report, generated and ready to forward
It arrives as a clickable link, live for seven days, printing to PDF with a client-ready filename — no renaming. And the recommended actions print as the final page by design: include it and the report is your proposal, or drop the last sheet and it’s a findings-only deliverable.
What’s in it
Ask for it by name
“Generate the white-label visibility report for Acme's last run and give me the link.”
Not a rank tracker with a new label
The engines are asked, not scraped — and the loop ends with a fix, not a score.
Asked, not scraped
Every question runs live with web search on — the same answering mode your client's buyers actually see. What comes back is the real answer, verbatim, not an index approximation.
Share of voice, named
Not just whether Acme appears, but who appears instead. The comparison runs against the competitor set your client names, so the gap is one they already care about.
The action tail
Trackers stop at the score. This lane ends with a corrected llms.txt written against the exact questions you're losing — the fix, not just the diagnosis.
The delta is the deliverable
Re-run the same locked pack next month and every number arrives with its movement attached. That comparison is what a retainer bills against.
Four playbooks, one loop
Run them in order the first month, then just re-run the measure and the report — the pack is already locked.
A fifty-question pack — drafted from the brand kit, reviewed by you lane by lane, then locked so every future run is a fair comparison.
Two hundred live answers scored — the shelf, the gaps, and the reputation read, with every number carrying its month-over-month movement from the second run on.
The corrected llms.txt — written against the measured gaps, improving the client's existing file rather than replacing it, with every change documented.
The white-labeled GEO audit report — executive verdict, four-engine scoreboard, share of voice, verbatim exhibits and the reputation chapter, generated and linked, ready to forward.
This is one lane. The playbook idea runs the whole agency.
Creative, intelligence, and account work — over 150 playbooks, all run by talking to Claude, all on your brand. AI Visibility Pulse is just what it looks like up close.