How GEO Results Are Measured (With a Sample Report)

GEO results are measured by one main number: how often an AI engine names your brand inside its answer. You ask the same buyer questions in each engine, from your buyer’s market, on a set date, and record whether your brand is named and where. Then you compare it with the month before. Rankings, clicks and source citations still go in the report, but they sit underneath that number, because none of them tells you whether the AI recommended you.

What is the main number in a GEO report?

The main number is the named rate. It answers one plain question: when a buyer asks this engine this question, does the answer name you?

Here is how it is counted. Pick the questions your buyers actually type, such as “best accounting firm for startups in Singapore”. Ask each one, logged out and from your buyer’s country, in Google AI Overviews, Google AI Mode, Gemini, ChatGPT and Claude. For each answer, record whether your brand is named, in what position, and who is named beside you.

Named in four of six checks on a question in ChatGPT means a named rate of four in six. That goes at the top of the report, one row per engine, against the month before.

What is the difference between being cited and being named?

Cited means your page is listed as one of the sources under the answer. Named means your brand is in the answer text itself, as one of the businesses the AI recommends. A buyer reads the answer. Few of them open the sources.

The two come apart more often than people expect. In August 2026 AeroChat, Alva’s own product, ranked third on Google for “best Shopify chatbot” but was not named in the AI Overview above the results. When we went through the full list of sources behind that answer, one article named AeroChat first, but none of the other sources mentioned it at all. So the AI didn’t either.

That points to what drives the named rate. In September 2026 we looked at 40 Google AI Overview results for a branding agency client. A business that appeared on one source was named 27% of the time. A business on three or more sources was named 87% of the time. Your own pages get you cited. Other people’s pages describing you the same way get you named.

A good report shows both: cited as the early sign, named as the result.

Why should each engine be measured on its own?

Because the engines disagree, and a single blended score hides it. We saw this clearly with Qwang Coffee, Alva’s own store. The domain was registered on 23 August 2026. It was indexed in one hour, on page one in 24 hours and cited in a Google AI Overview in 48 hours. It was listed first on 3 of 5 commercial questions we checked in Singapore.

A single blended score would have shown a small dip. The real story was a full loss on one engine and a strong hold on another.

So the report has one row per engine. If an agency gives you one number for “AI search”, ask to see it broken out.

Why do the market, the language and the date matter?

Because the same question gets a different answer depending on where it is asked, in what language and on what day. A result without those four labels (engine, question, market, date) can’t be checked or compared.

Two of our own checks from November 2026 show how big the difference is:

• Market. On 2 November 2026 we asked 15 buyer questions from a Malaysian location across five AI engines, 270 answers in all. The answers leaned on Malaysian directories, local “best of” articles and firms with a clear Malaysia service page. Stridec is a Singapore agency, and the answers asked from Malaysia mostly named Malaysian agencies. If we had only checked from Singapore, we would not have seen that.
• Language. On 4 November 2026 we asked “best GEO agency” in English and in the local language from Vietnam and from Thailand, in Google AI Mode. In Vietnam only one agency appeared in both the English and the local-language answer. In Thailand, none did. The same buyer, asking in a different language, got a different shortlist.

Dates matter too. Answers change week to week, so an old screenshot tells you what the engine said then, not now.

Why does traffic alone mislead?

Traffic misleads because an AI answer can name you and send no click, and because clicks move for reasons that have nothing to do with AI.

Our own Search Console data shows the first problem. On stridec.com, from June to August 2026, pages in positions one to three for non-branded searches had a click-through rate of 0.08%. Branded searches had 27%. People saw us high in the results and mostly didn’t click. The branded figure is consistent with people meeting us first and searching for us by name later. If you only watch non-branded clicks, you miss that.

Branded search is often where being named shows up in your traffic. For AeroChat in early 2026, period over period, impressions grew 343%, clicks grew 127% and branded search reached a 74% share. Those numbers are consistent with buyers meeting the brand in AI answers and then looking for it. They don’t prove it on their own, which is why the named rate sits above them.

Counts can also move in opposite directions. Bing’s Copilot data for stridec.com, read on 7 November 2026, showed our share when cited held at about 20 to 24%, while the number of questions citing us fell. A report with only one of those numbers would tell you either “steady” or “falling”. You need both to know what happened.

What does a GEO report look like? A sample

A useful GEO report opens with the verdict, then the numbers, then the reason. In order:

1. Verdict in one line: improving, stagnant or dropping.
2. Named rate now against last month, one row per engine, with the change.
3. What changed and what caused it: which piece of our work, or which move in the market.
4. What to do next.

Below is what the first two parts look like. This is an illustrative example. The brand and every figure are invented to show the format. It is not client data.

And the detail behind it, per question. Again, illustrative example, invented figures:

Read it the way an owner would. The verdict is improving: named more often on four of five engines. Gemini hasn’t moved, so that is where the next month’s work goes. The “Cited, not named” row is the one to watch: the engine had found the firm’s page but wasn’t yet recommending the firm. Traffic, rankings and branded search sit below all this as supporting numbers.

How often should GEO results be checked?

Check on the same day each month, with the same questions, from the same location, so each month is a fair comparison with the last. If you add questions, report them separately. And keep the misses in, next to the wins. A report that only shows wins can’t tell you what to fix.

What should you ask an agency about its GEO reporting?

Ask for a sample report before you sign. Check that it shows the named rate per engine, labels every result with engine, question, market and date, separates cited from named, and compares like with like. Our lists of questions to ask an SEO agency and red flags when hiring an SEO agency both include a GEO section with more of these.

What to do next

If you just want a first reading, start small. Our guide to checking whether you show up in ChatGPT in 15 minutes walks through a quick version of this method. If you are deciding who should do the work, hire an agency, train your team, or do it yourself sets out the honest fit for each route. For a deeper method on tracking, see how to measure AI search visibility.

Frequently asked questions