If an agency sends you a screenshot of ChatGPT recommending your business, ask for the prompt. Did it ask for an agency with your capabilities, or did it name your company in the question? Those are very different tests.
Keep the screenshot. Just make sure the report explains what it actually demonstrates.
To measure GEO, track four things separately: whether your business appears in relevant AI answers, whether those answers cite your pages, whether people visit, and whether those visits or reported discoveries lead to qualified inquiries. A mention is evidence of visibility in that response. It is not evidence of a sale.
GEO, or generative engine optimization, is work intended to improve how a business is found and represented in AI-assisted search. The reporting should help you decide what to improve next, not just collect flattering answers.
Start with questions a buyer would actually ask
Build a small, stable set of questions from sales conversations and customer research. Include the service, the type of business, and any constraint that changes the recommendation.
For an ecommerce agency, “Who can help a skincare brand manage TikTok Shop creators and paid ads?” is more useful than “Tell me why Best Practice Media is great.” The second prompt supplies the conclusion you are hoping to measure.
Keep unbranded discovery questions separate from branded due-diligence questions. “Which agencies support ecommerce paid social?” tests something different from “What services does Best Practice Media offer?” Both matter, but combining them into one score can conceal a discovery gap.
Choose a manageable cadence, such as a weekly check. Save the exact question and response, the date, the product or engine, any visible model setting, location where relevant, and whether the session was fresh or carried prior conversation. Answers can vary. A controlled sample is more useful than rerunning a question until you like the result.
A GEO measurement template that keeps the evidence straight
Use one record per question and observed response. Keep the following fields alongside a separate traffic and inquiry report.
| Measure | Record | What it can tell you |
|---|---|---|
| Brand presence | Whether your brand appears; recommendation, incidental mention, or another context | Visibility within this specific sample |
| Accuracy | Services, location, fit, and any incorrect claim, with the exact wording | Whether buyers are getting a useful description |
| Citations | Exact linked URL and the claim it supports; distinguish BPM pages from third-party pages | Which sources the response visibly points to |
| Referrals | Observed source, landing page, reporting period, and relevant site actions | Measurable visits that carry identifiable source information |
| Qualified inquiries | Service fit, source evidence, self-reported discovery, and lead stage | Commercial relevance, with attribution limits preserved |
| Next action | Specific page or factual gap, owner, and review date | What the team will change based on the findings |
For a simple visibility measure, divide responses that mention the brand by all completed responses in the same predefined sample. Report the raw count too. If six of 20 responses mention you, that is 30% of this sample—not 30% of all AI searches or buyers.
That example is hypothetical. Record failed or unavailable checks separately rather than quietly removing unfavorable observations. If you change the questions, tools, or sampling method, start a new comparison series or clearly flag the break.
Do not turn every Google visit into an AI visit
Google says traffic from its AI features is included in Search Console’s overall Web search performance reporting. That means an increase in ordinary Search Console clicks, by itself, does not prove that AI Overviews or AI Mode caused the increase.
Keep Google organic performance in view, but label what you can actually observe. For other AI products, record identifiable referrals when your analytics receives them. A missing referrer, a later direct visit, or a buyer switching devices can leave the journey incomplete. “No measured AI referrals” is a narrower statement than “AI sent nobody.”
Make sure a completed inquiry is measured as a completed inquiry, not merely a click on the contact button. Reconcile analytics with the actual leads your team receives. Record source information according to your consent and privacy practices.
A short “How did you hear about us?” field can add context. Keep that answer beside the analytics source instead of replacing it. Someone may first see your name in an AI answer, then search Google and submit a form days later.
Ask for these deliverables before signing a GEO proposal
A promise to “increase AI visibility” leaves a lot undefined. Ask the agency to show you what you will actually receive and what your team will need to supply.
- A starting benchmark: the buyer questions, engines, observation conditions, and saved answers used for comparison. Include incorrect answers and missed mentions.
- A page-level work plan: which service pages or resources need changes, what evidence is missing, and who approves factual claims. Publishing more articles should not be the automatic answer.
- A source record: where client results, quotes, and business details came from. Distinguish changes the agency can make on your site from third-party coverage it can only pursue.
- An implementation owner: who handles copy, technical changes, analytics, and approval. A recommendation in a slide deck is different from a verified change on the website.
- A review that leads to a decision: what changed, what was observed, what remains uncertain, and which action comes next. Keep qualified inquiries beside the visibility measures.
Ask for a sample report with confidential information removed. You should be able to trace one conclusion back to its evidence. If you cannot, a higher visibility score will be hard to evaluate too.
Use the report to decide which page needs work
If you appear but the service description is wrong, review the pages and third-party sources named in the response. Correct your own outdated copy. Record external inaccuracies for appropriate follow-up rather than assuming a single website edit will immediately change every answer.
If buyers land on a useful article but never explore a service, check whether the next step is obvious and relevant. If inquiries arrive but the fit is poor, tighten who the service is for and what it includes. Those are different problems from simply needing more mentions.
Google’s guidance also makes clear that its AI features do not require special schema or a separate set of technical optimizations. Helpful content, crawlable pages, appropriate internal links, and accurate structured data remain part of the work. Eligibility is not a promise of inclusion.
Our GEO services focus on helping buyers find and understand BPM clients through AI-assisted discovery. That work connects with organic search strategy; reporting should preserve the distinction between observed visibility and attributable business results.
Paid placements are a separate channel. Keep any ChatGPT advertising spend and results separate from organic recommendations.
At the next review, put one observed answer beside the page it cites and the inquiries received that month. You may not be able to connect them all. Say where the evidence stops, then choose one specific improvement to test. That is a report you can make a decision from.
