When a client asks for an AI visibility report

The ask now comes from the client.

More of a client's buyers now ask an AI tool before they search. Ofcom reported in April 2026 that 54% of UK adults aged 16 and over use AI tools, from its survey of 7,533 adults in autumn 2025. Google said in May 2026 that AI Mode had passed 1 billion monthly users worldwide, a figure that's its own and not independently audited.

So the question lands on the agency: are we showing up in ChatGPT, and can you report on it? You can, if the report says what it proves and what it doesn't, names who owns each number, and is priced on work the agency controls.

Say what the report proves before you sell it.

An AI visibility report runs a fixed panel of prompts through the AI platforms the client's buyers use and records what comes back. It shows whether the brand appears, how often across repeat runs, which competitors share the answers and which pages get cited. It can't show market share, influence or revenue.

Clicks are rare. Pew Research Center found that when a Google results page showed an AI summary, a link cited in it was the next visit 1% of the time, across 68,879 searches by 900 US adults in March 2025. The visits that do come can be worth more. Adobe found AI-referred visitors to US retail sites converted at a 60% higher rate than other traffic in July 2026, under its own method. Visibility and visits need separate lines in the report.

Agree who owns each number.

The report crosses 3 teams, and the agency usually holds only the first layer. Write the split into the scope, or the agency ends up answering for revenue it can't see.

Scope it so the price holds.

Price the panel and the reading of it, not the tool. A checker's licence is a cost to pass through. The work is choosing 30 to 50 real buyer questions, running each more than once on each platform, checking the brand is matched correctly, and explaining each month what changed and what the agency will do about it.

Scope it on what drives the hours: how many prompts, which platforms, how many repeat runs, how many competitors and how often. Split month 1, when the panel is built and each of the 4 layers gets a baseline, from the monthly runs after it. And keep it out of the SEO retainer as an unpriced extra. It's separate work with its own owner.

Don't price on a score target. A rising score is a rising sample, and Bing calls its own Citation Share metric observational, not a ranking or a competitive scoreboard. A fee tied to a number nobody controls starts an argument the agency can't win.

What not to promise.

Don't promise a share of the market, because the panel is a sample. Don't promise that citations made sales: a citation shows a source fed an answer, not that the answer sold anything. Don't promise the score will rise every month. And don't hide the method: a credible report shows the panel, its timing, the sample size and its limits.

Promise a documented panel, a baseline across the 4 layers by the end of month 1, and a monthly read of what changed. That's a report a client can take to its board.

The method is in Dataflos's article.

This read covers the client conversation. The method, including 5 tests for any AI visibility checker and the full 4-layer measurement, is in Dataflos's article on what AI visibility checkers prove. Dataflos is Twist Media Group's data team.

For agencies, our data and measurement service builds this reporting for a client account. For brands, data and strategy builds it in-house.