How agencies report AI visibility to clients

Because their buyers now get answers without clicking. When a Google results page shows an AI summary, Pew Research found that users clicked a traditional result on 8% of visits, against 15% when there was no summary (Pew Research Center). Clients see flat traffic and ask whether they're losing ground somewhere they can't see.

An AI visibility report answers that question. It shows whether ChatGPT, Gemini, Perplexity and other assistants mention the client, how often compared with competitors, and why. For an agency, it is also the clearest proof that GEO work is doing something, which is what keeps a retainer alive.

One page of findings, then the detail for anyone who wants it. Busy clients read the first page; their analysts read the rest.

Page one:

1. One headline sentence. "You now appear in 31% of category answers in ChatGPT, up from 24%. Competitor A still leads at 42%."
2. Share of voice per engine, against three to five competitors the client recognizes.
3. Prompts won and lost this month, with the most important ones named.
4. What we did and what we'll do next, in three bullets each.

Detail pages:

• mention rate and average position per engine and market;
• the sources cited most often for the client's prompts, and which ones mention them;
• how engines describe the client, with example sentences;
• the method: prompt count, engines, markets, run frequency and dates.

The metric definitions are in our AI visibility metrics guide, and the share of voice calculation is in our guide to measuring AI share of voice.

The first report is a baseline, and it sets the tone for the whole relationship. It has no trend yet, so it has to earn attention with the gap.

Structure it around three questions the client is already asking:

1. Are we mentioned? Mention rate per engine across the agreed prompts, with two or three example answers quoted in full.
2. Who is mentioned instead? The competitor leaderboard for the client's category, per engine and market.
3. Why? The ten sources cited most often for those prompts, marked by whether they mention the client.

Close with the plan for the first 90 days: the technical fixes, the pages to rewrite and the third-party sources to work on, each with an owner. A baseline report that ends with a plan turns a worrying number into a project the client can approve.

Agree on the prompt set before you send it. If the client's team helped choose the questions, they won't dismiss the results as "not the prompts our buyers use."

Compare it with rankings, then show one real answer. Rankings tell you where a page appears in a list of links. AI visibility tells you whether an assistant names the brand inside its answer, and in what order.

Most clients get it the moment they see a specific prompt. Open the stored answer for "best [their category] for [their audience]", point to the three brands the assistant recommends, and show that theirs isn't one of them, or is listed last. Then show the cited sources: a review site that doesn't list them, a comparison article from two years ago. The problem and the fix become concrete in a minute.

Keep the link to SEO clear. Google's AI features use its search index (Google Search Central), and many assistants search the web, so good SEO still matters. AI visibility adds a new question on top: once you're found, are you recommended?

With stored answers, not opinions. When share of voice moves, open the prompts that changed and look at what the engine said and cited, before and after.

Common explanations, in the order to check them:

• The competitor set changed. A new brand added to tracking lowers everyone's share.
• The engine changed. A model update can shift many answers at once.
• A source changed. A review site or listicle added or dropped the client.
• Our work landed. A new comparison page or a fixed pricing page started being cited.

Name the cause in the report. "Down 4 points because Perplexity stopped citing the 2024 roundup that listed you; we've pitched the 2026 edition" is a sentence a client can accept. "Fluctuation" is not.

Honest ones. No agency and no tool can guarantee what an AI engine will say, because engines decide their own answers and change their models and sources over time. Say so in the proposal.

What you can promise:

• a stable, repeatable measurement of the client's visibility against competitors;
• a clear diagnosis of why competitors win specific prompts;
• a prioritized plan, carried out on schedule;
• monthly evidence of what changed and why.

Set the timeline too. Engines that search the web live, such as Perplexity, can reflect changes within weeks. Answers drawn from model memory move only when new models are trained, which can take months. A first quarter usually goes on baseline, technical fixes and source work, with clearer movement after that.

Monthly for the client report, with a deeper review every quarter. Weekly reporting invites clients to react to normal noise; answers vary from run to run, and a weekly dip often reverses on its own.

Match the cadence to the client's own reporting rhythm. If their marketing team reports to the board quarterly, make sure your quarterly review lands a week before, with the numbers in a form they can paste into their own deck. A report that saves the client work is the one that gets read.

Track daily, though. Daily runs give you a smooth trend and let you spot real drops early, so you can tell the client about a problem before they find it. Share a live link for clients who like to look in between reports, and keep the monthly PDF as the record.

Most agencies add it in three layers, priced by the number of prompts, engines and markets:

1. Audit. A one-off baseline: where the client stands, who wins instead and why. It doubles as the pitch.
2. Monitoring and reporting. Monthly tracking and the report described above.
3. Optimization. The work itself: content fixes, comparison pages, review profiles and outreach to cited sources.

Keep the prompt set agreed in writing, and review it each quarter. It is the client's definition of success, so changing it quietly undermines every trend line. Our page on Ossvy for agencies covers multi-client setup and per-client usage.

The common ones:

• Screenshots instead of data. One chat screenshot proves nothing about the next run.
• Blended global scores. A strong home market hides a weak export market. Report each market.
• Changing prompts without saying so. Every trend breaks.
• Taking credit for every rise. Some changes come from model updates. Clients trust agencies that say so.
• Numbers without answers. A share of voice chart feels abstract on its own. Quote the one answer that changed most this month, before and after, so the client sees what the number means.
• Jargon. "Citation rate in grounded responses" means nothing to a CMO. "How often ChatGPT links to your site" does.

Report AI visibility on one page: share of voice per engine against known competitors, prompts won and lost, sources, actions done and next. Track daily, report monthly, review quarterly, and keep the prompt set agreed in writing. Explain every change with stored answers, and set honest expectations from day one. Ossvy's white-label reports build this report under your logo, and the free AI visibility check makes a quick first look for a prospect.