Case study — custom engagement
A B2B SaaS platform was earning thousands of visits a month from ChatGPT, Claude, Gemini and Perplexity. Those visitors signed up at five times the rate of everyone else — and the pages receiving most of them had no way to sign up at all. We triaged all 429 of them.
This is a real engagement, anonymized. The client is called Gridwell here, every absolute figure is scaled by a constant, and the vendors and internal systems named in the report are renamed. The findings, the method and the shape of the data are the ones we delivered.
Assistants sent a trickle for nine months, then stepped up sharply in May and held. Conversion rate never dropped through that change, so the extra volume arrived at the same quality.
The same visitors, arriving from the same assistants, converting between 6.6 and 29.1 percent depending only on which page they landed on. That spread is the whole opportunity.
Ranked by accounts recoverable, not by how bad they look. Each was checked against the client's own conversion data before it was recommended.
Assistants sent 864 visits to help center pages that answer a question and then stop. Thirty-three accounts came back. The same visitors convert at 20.4 percent elsewhere on the site.
Privacy and terms drew 619 visits from assistants answering questions about security and data handling. Those pages carry no route into the product at all, and produced four accounts between them.
View Builder took 252 visits last month, up 73 percent, the fastest growth of any theme in the report. Automations took 230, up 32 percent. Both convert at roughly half the site's rate for AI traffic.
Sessions arriving from assistant domains, isolated in the client's own warehouse and credited to the page the visit started on. Demo requests are the product's account-creation event, not an analytics estimate.
All 429 pages that earned AI traffic, sorted by the accounts they should have produced at the site's own rate and did not. Volume alone would have put the homepage first and hidden the actual problem.
The client's AI-visibility vendor had a recommendation backlog. Each item was matched to the pages it targeted and judged on what those pages actually converted. Some were adopted; others were not worth the effort.
A tracking gap, a modeled demand column that must never be read as a ratio against Google volume, and a one-snapshot caveat. Written into the report so no one builds a plan on a number that cannot carry it.
The client received a single self-contained report: summary, priority actions, traffic over time, conversion rate by category, the pages losing the most, pages competing with each other, the pages already working, the vendor queue checked against conversion data, and the method.
The full anonymized version is published — charts live, tables sortable, every figure the same one the client saw at a constant scale.
| Search | Page that should own it | Rate |
|---|---|---|
| what is gridwell | /guides/start/what-is-gridwell |
30.2% |
| no code ai tools | /articles/no-code-ai-tools |
5.6% |
| is gridwell free | /pricing |
28.1% |
| ai agent builder | /articles/best-ai-agent-builders |
— |
| gridwell cost | /pricing |
28.1% |
If assistants are already sending people to your product, the same analysis applies. It needs your analytics, your demo request event, and about two weeks.