
An agency can offer AI visibility reports as a measured service line: track a fixed set of each client's buying questions, ask each one several times across AI tools and languages, and report how often the client is named and which sources the AI cited. Then turn what you find into work your team can price.
The common mistake is trusting a single answer. In our Malaysian test of ChatGPT, fewer than one in ten of the businesses it named appeared every time we repeated the same question.
So the report your clients need isn't a screenshot. It's a rate, a source map and a list of fixes.
What should an AI visibility report measure for my agency's clients?
An AI visibility report for an agency client should measure five things: how often the client is named across repeated questions, whether the answer links to the client's own page, which sources the AI cited instead, which competitors took the answer, and whether the AI describes the client correctly. Each points to a different kind of work.
| Measure | What it tells the client | Work it points to |
|---|---|---|
| Mention rate: share of answers that name the client | How visible they are for their buying questions | Pages and listings for the questions they lose |
| Citation: the answer links to the client's page | Whether the AI used the client's own page | Plain-text facts on the pages that should be cited |
| Sources cited instead | Where the AI got its answer | Listings, directories and third-party pages |
| Competitors named | Who holds the answer today | A page that answers that question better |
| Accuracy of the description | Wrong prices, hours or services | Corrections at the source the AI cited |
Don't report a rank. In a SparkToro study, volunteers ran 12 prompts a combined 2,961 times. Rand Fishkin of SparkToro put the lesson this way: "visibility % across dozens to hundreds of prompts run multiple times is a reasonable metric"1. A position number in a single answer is not.
To see an AI's full set of recommendations, Fishkin says you'd ask the same question "usually at least 60-100X"1. A small panel gives a rough baseline, not that full picture.
Clients who only know Google rankings need this explained up front: in an AI answer the client is named or not, and the list changes from one asking to the next.
Which questions should my agency track in a client's AI visibility report?
Track the questions a client's customers ask right before they buy: who to hire nearby, what it costs, and facts about the client by name. In our test, ChatGPT searched the web every time for all three kinds:
- Who to hire nearby: 18 of 18 answers.
- Prices: 12 of 12 answers.
- A well-known business by name: 12 of 12 answers.
General questions behave differently. In the same search-or-memory data, ChatGPT often skipped the search:
- Definitions: it searched for only 5 of 15 answers.
- How-to questions: it searched for 8 of 15.
Every answer it gave without searching came with no sources. Questions like those mostly measure what the model already knows.
A practical panel is about ten questions per client, written the way customers actually type them into ChatGPT. For an aircon servicing client, that might be "Can you recommend a reliable aircon servicing company near Puchong?" Keep the wording fixed each month.
Should my agency's AI visibility reports cover Malay and Chinese questions?
Cover Malay and Chinese questions if the client's customers ask in them, and report each language separately. In our language test, ChatGPT mostly searched in English, whatever language it was asked in:
- Asked in Chinese: 96% of its web searches were in English.
- Asked in Malay: 87% were in English.
How does my agency run AI visibility checks honestly for clients?
An agency runs AI visibility checks honestly by asking each question at least three times, recording the date, AI tool, language and location, and using no chat history. AI answers vary from run to run. In our Malaysian test, two answers to the same question in the same language shared only 15% of the businesses they named on average, though hotel answers were far steadier. Three runs is a minimum for a rough baseline, not proof.
Ask while logged out when you check by hand. OpenAI says signed-out users can use ChatGPT's web search, and that with memory on, ChatGPT "may use relevant saved memories when rewriting a search query"2. Logged out, answers aren't shaped by your team's own memories and history.
| One-off check | Repeated panel | |
|---|---|---|
| Runs per question | One | Three or more, every cycle |
| What you can report | "ChatGPT named you once today" | The share of answers naming the client, and its trend |
| Month-to-month change | Can't be separated from noise | Small moves are still noise; add questions or runs before claiming a trend |
How should my agency present AI visibility results to clients?
Present AI visibility results as three layers: the mention rate and its trend, a map of the sources the AI cited, and a short list of fixes tied to specific questions. The source map shows where the answer came from, so it shows where the work goes.
| Question type | Largest source ChatGPT cited | What the agency fixes |
|---|---|---|
| Who to hire nearby | Businesses' own sites, 59%. Next: directories, marketplaces and apps, 27% | The client's service pages, then their directory listings |
| A price | Businesses' own pages, 58%. Next: directories, marketplaces and apps, 32% | A published price or cost guide on the client's site, and their directory listings |
| A time-sensitive fact | Government sites, 82% | Little: point to the official source, don't compete with it |
Every cited site, by question: download the CSV.
Map sources per client trade, not from an average. In the same data, single questions swung the totals:
- Wedding photographer: this one question gave 14 of the 30 own-site citations for local questions.
- Renovation: this question drew three directory citations and only one from a business's own site.
- Website cost: this question gave 8 of the 11 own-page price citations. Without it, directories lead for prices.
Add two measures clients already trust. OpenAI says ChatGPT "automatically includes the UTM parameter utm_source=chatgpt.com in referral URLs"3, so referral visits show up in the client's analytics. For Google, its guide points to "the Generative AI performance report in Search Console"4.
What should my agency never promise in an AI visibility report?
Never promise a client a place in AI answers, a rank, or a result caused by one change. OpenAI says of ChatGPT search that "Placement is not guaranteed"2. Google's guide is blunter about tools: "Be wary of third-party tools that promise ranking success or claim to use "internal" Google metrics. No third-party tool has access to our internal ranking or AI systems."4
What an agency can promise is the work and the measurement: which questions, how often, on which AI tools and languages, and what it will change. Report a rise as "the mention rate moved from X to Y after these changes", not "this change got you cited", because one month's move may be noise.
What can't an AI visibility report tell my agency's clients?
An AI visibility report can't show exactly what a particular customer saw, and it can't prove why the AI chose a source. Tell clients these limits up front:
- What one customer saw. The ChatGPT app can use a person's location, saved memories and chat history. A controlled check leaves those out.
- Why a source won. A check shows only what was cited. Even the sources an API lists as consulted don't show why one won, so label any "why" as your inference.
- How far our figures reach. Our own figures came from OpenAI's API, not the ChatGPT app. They cover one engine and one set of Malaysian questions, with three runs each.
How can my agency start offering AI visibility reports this month?
Start by getting a free check on one client from EasyFound, then build the report around what it shows. The opening is real: in our test, ChatGPT looked up every local recommendation on the web, and businesses' own sites were the largest group of sources it cited, so the fixes are work your agency already does.
- Send one client's website through easyfound.co/agencies and EasyFound will email you what AI says about that client, or book the free 30-minute walkthrough on the same page. No sign-up or card is needed.
- Write the client's panel: about ten buying questions in the words their customers use, in each language they're asked in.
- Run each question three times and map the sources. A single answer settles nothing, because the names change from run to run. Budget for the reading: ten questions, three runs and three AI tools is about 90 answers per client each cycle.
- Pitch the report together with the first three fixes, as a baseline audit or an add-on to the client's retainer. A report on its own is what self-serve monitoring tools already sell.
For agencies, EasyFound, an AI search visibility (GEO) tool built for Malaysian businesses, tracks what six AI engines say about each client in English, Malay, Mandarin and Manglish, with your clients on one agency account. The six are ChatGPT, Claude, Perplexity, Gemini, Google AI Overviews and Google AI Mode. ChatGPT, Claude and Perplexity show by default, and EasyFound can switch Google's three on for an account.
White-label reports and automatic fixing aren't available yet, so your agency does the fixing work. Start with one client below.
Sources
Our data: EasyFound, three tests of ChatGPT through OpenAI's Responses API (model gpt-5.4-mini, web search available, user location Malaysia, no chat history), September 2026: a language test of 45 questions × 3 runs in English, Bahasa Malaysia and Chinese across five industries (24 September; write-up, data); a search-or-memory test of 30 questions × 3 runs (26 September; data); and the sites cited for those answers, by question type (data). Market prices were read from Malaysian providers' and tools' own pricing pages on 26 September 2026, with the lowest and highest rechecked on 28 September 2026. We don't name the businesses or providers, because one test isn't a ranking.
Other sources:
- SparkToro (Rand Fishkin and Patrick O'Donnell), NEW Research: AIs are highly inconsistent when recommending brands or products; marketers should take care when tracking AI visibility, January 2026 (link)
- OpenAI, Searching the web with ChatGPT, read 28 September 2026 (link)
- OpenAI, Publishers and developers FAQ, read 28 September 2026 (link)
- Google Search Central, Google's Guide to Optimizing for Generative AI Features on Google Search, updated July 2026 (link)
AI answers and published prices change often; treat these figures as dated measurements.
Research by Janice Pang, EasyFound.

Is there an AI SEO tool where someone does the fixing for you? (Malaysia)
Which type of AI visibility or GEO tool is best for a small business in 2026? (Malaysia)
Is one ChatGPT screenshot enough to report our AI visibility? (Malaysia)