Quarterly AI visibility reviews for clients
How we run a quarterly review of a client's AI search program: the data window, the period comparison, the exports we keep, and the report that goes out afterwards.
Monthly reports tell a client what moved. Quarterly reviews tell them whether the program is working. We run both, but the quarterly meeting is where budgets get defended or changed, so we prepare for it differently.
Here is the routine we use on client programs, and the Promptwatch features that make each step less manual.
Why a quarter, and why the date range matters
AI answers are noisy from one run to the next. A single week can move because of a model update, a news cycle, or one popular page being re-crawled. Three months smooths enough of that to show a direction without hiding it.
There is a practical side too. The Promptwatch REST API v2 and the MCP server accept date ranges up to 92 days. A calendar quarter is 90 to 92 days, so one request covers the whole window. We pull each quarter in a single pass with fixed start and end dates, and we write those dates into the review document. That sounds trivial until a client asks, six months later, exactly which period a chart covered.
Step one: freeze the panel
Before we pull anything, we check that the prompt panel did not change during the quarter. If prompts were added or removed, the comparison between periods is partly a comparison between different questions. We note every change from our own log and decide whether the affected prompts should be excluded or shown separately.
We do not hide this from the client. A panel change is part of the story, and it is better explained in the meeting than discovered afterwards.
Step two: compare the period to the one before
Promptwatch has an A/B period comparison, which puts two date ranges side by side under the same filters. We compare the quarter to the previous one, and where the program is old enough, to the same quarter a year earlier.
What we look for is not the headline number. We look at which topics, models, and competitors changed. A flat overall result often hides one engine that improved and another that slipped. The comparison view makes that visible without building the same chart twice.
We keep our interpretation careful. A movement is something to investigate, not proof that a specific piece of work caused it. We say so in the review.
Step three: pull the raw responses
The summary charts answer "what happened". The responses answer "what did the engine actually say". We export the responses, which since November 2 lists each brand mention with its position in the answer, for example Acme (1), Globex (2). That makes it simple to filter a spreadsheet down to the answers where a client appeared late, or where a competitor took first place.
We read a sample of those answers ourselves before the meeting. A client will always ask about one specific answer they saw, and it helps to have read it first.
Step four: connect it to the client's own reporting
Most clients already have a reporting setup, and they do not want a second one to learn. Promptwatch offers custom reports and a Looker Studio connector, with templates, so the AI visibility data can sit next to their search and traffic numbers. On Professional and above, and on all agency plans, that connector is part of the package.
For clients who prefer a document, scheduled PDF reports go out on a cadence. A scheduled custom dashboard follows its widgets and name as they change, so a quarterly pack does not need rebuilding by hand. We read it before it is sent, every time.
Step five: structure the meeting
Our quarterly review follows the same order for every client.
- What we agreed to track and what changed in the panel. Two minutes, no more.
- What moved, by topic and by engine. The period comparison, with the interesting segments pulled out.
- What the answers actually said. Three or four real responses, chosen to show a pattern.
- What drove traffic and conversions. Visitor analytics next to the visibility data, so the room can see whether visits followed.
- What we recommend next quarter. Two or three actions at most, each tied to something observed.
We leave out any slide that exists only to look busy.
What we do not promise
We do not promise that a good quarter will repeat, and we do not claim a specific action produced a specific change in the numbers. The data shows what the engines said in a window. It does not isolate causes. Clients take that better than we expected, especially when we show our reasoning and the raw answers behind it.
Keep the quarter's files
After the meeting we archive the exported responses, the report, and a short note listing the date range, the panel version, and any caveats. When the next review arrives, we do not have to reconstruct what we did last time. That archive has saved more arguments than any chart.
If you would like to try the period comparison and the exports on your own data, Promptwatch has a 7-day free trial. If you would rather have us run the quarterly cycle, write to hello@1001seomedia.com.