Citino teamAgencies track AI visibility for clients the same way a brand tracks its own, multiplied by the number of clients: a baseline before the work, a fixed set of questions and competitors per client, daily answers from ChatGPT, Google AI Overviews and Perplexity, and a monthly report with a verdict on every fix. The hard part is keeping each client’s numbers honest and comparable over time.
The measurement itself is AI search analytics. This playbook covers running it across a roster.
Step 1: Run a baseline before the work starts
Measure where the client stands before you change anything. Without a baseline, you cannot show what your work moved.
- Agree the questions the client’s buyers ask, worded as buyers word them.
- Agree the competitors, usually three to five that already appear in answers to those questions.
- Collect at least two weeks of daily answers on each engine before the first fix ships, so the starting numbers have a usable margin of error.
The baseline is also a strong pitch: showing a prospect where they stand against their competitors, before they sign, makes the case for the work.
Step 2: Build a prompt set for each client
Each client needs its own prompt set, built the same way you would build one for your own brand. How to choose the prompts to track covers the method. For agencies, three rules matter most:
- Group prompts by topic, and size the groups by what you report on. If the retainer is about billing questions, put enough prompts in the billing group to measure it, not one.
- Keep branded prompts separate. A question that names the client almost always gets an answer that names the client.
- Never reword a live prompt. Language models are sensitive to small changes in wording (Sclar et al.), so a reworded prompt starts a new series. Retire it and add a new one.
Worked example (illustrative). An agency has a pool of 250 prompts across six clients. Two larger clients with several product lines get 70 each; four smaller ones get 25 to 30 each. The split follows what each retainer covers, not an equal share.
Step 3: Get the access you need
Two client accounts make the work faster and the reports more credible:
| Access | What it gives you | Who grants it |
|---|---|---|
| Google Search Console | Whether the client’s pages are indexed and eligible to appear | A property owner |
| Google Analytics | Visits from AI assistants and what they convert to | A user with the Administrator role |
Ask for read-only access where you can: a viewer role in Analytics and a restricted user in Search Console cover tracking.
Step 4: Measure every client the same way
Comparability is what makes a roster manageable. For every client:
- The same engines: at least ChatGPT, Google AI Overviews and Perplexity.
- The same schedule: daily, at a fixed time, from a clean session.
- The same metrics: visibility, share of voice, position and sentiment, each per engine with a margin of error.
If one client is measured weekly and another daily, or one with three competitors and another with eight, their numbers cannot be compared, and neither can your results across the roster.
Step 5: Tie every fix to a verdict
Clients pay for results, so each fix needs a verdict, not a screenshot. Before a fix ships, record the questions it targets. After an engine has read the changed page, measure those questions against the median competitor over the same days and compare the result with its margin. The verdict is Confirmed, Negative, Inconclusive, Measuring or Awaiting first retrieval. How to tell whether an AI search fix worked sets out the steps.
Step 6: Report monthly, in the client’s terms
Use the one-page structure in how to report AI visibility to leadership: a headline, share of voice against the client’s competitors, what moved and why, fixes and verdicts, visits from AI, and next month’s actions.
Two habits keep clients trusting the reports:
- Report flat months as flat. If nothing moved beyond its margin, say so, and show what is still awaiting retrieval.
- Never edit a verdict. A negative or inconclusive result reported plainly is what makes a confirmed one believable.
Mistakes agencies make
| Mistake | What it costs |
|---|---|
| No baseline before the work | Nothing to prove the work against |
| One prompt set shared across clients | Each client’s numbers measure questions they do not care about |
| Rewording prompts to “improve” them | Breaks the trend the client is paying to see |
| Screenshots of single answers as proof | The next screenshot can say the opposite |
| Reporting noise as wins | The first flat quarter ends the retainer |
The baseline you take before the first fix is the most valuable number in the account.
Where Citino helps
Citino is an AI search analytics tool. It asks ChatGPT, Google AI Overviews and Perplexity your clients’ buyers’ questions every day, with Gemini, Google AI Mode and Claude as add-ons, and reports every number with a 95% margin of error, as the methodology sets out. The Agency plan shares one prompt pool across unlimited client brands, with white-label reports, a portal for each client and a verdict on every fix. See Citino for agencies.