How do we measure GEO results, and what do we offer if they don't improve?

Every engagement starts with a baseline. We agree on 30–50 buyer questions per language, including relevant city and price constraints, and run each several times in clean, logged-out sessions across ChatGPT, Perplexity and Gemini. We save screenshots and share the full question set and initial results with you.

A fixed question set makes later comparisons meaningful. A dashboard showing +300% visibility says little if the questions or counting rules changed along the way. You keep the original list, so you can check every later report against it. This is the clean-session method used in our published case study: six runs, two machines, a VPN and no login.

Citation rate is the share of test questions where an assistant links to one of your pages, measured separately by assistant and language. A question counts if the link appears in at least one of its repeat runs; how consistently it appears is reported separately. Mentions of your brand without a link count towards share of voice, not citation rate. Share of voice compares your appearances with those of competitors named by the assistants; we record that competitor set at baseline. Language consistency checks whether answers describe your business the same way in each language you use.

Different tools can give the same company different visibility scores because answers vary between runs, question phrasings and regions. We therefore repeat the same question set and share the raw logs. As a final check, we compare visibility with visits from assistant domains in your own analytics. You retain control of those data.

Each month, we repeat the original questions several times per assistant and compare the results with the baseline, separately by language and assistant. Your report lists completed work on content, company information and technical issues. It also includes next month's plan and explains why we recommend each action. You can inspect the logs behind the report.

The raw records are included: the question list, saved answers and screenshots. You can repeat any question in a logged-out session and compare the result with the logs. The method is documented so you can check how we arrived at the reported figures. That is not yet the norm: in our September 2026 check of the 60 GEO agencies that ChatGPT, Perplexity and Google AI Mode cited in our runs of 14–15 September, 13 (22%) described their measurement method in full and 1 of 60 published raw run data. Our method is on this page, and our own runs are documented in our case study.

GEO work develops over two to three months, so we review progress at day 90. If citation rate and share of voice have not improved from the baseline after 90 days, we tell you and explain what we found. You choose a free diagnostic month, a change of scope or an end to the engagement. You keep your data without a penalty for leaving.

What we do not promise, at day 90 or ever: a specific answer from a specific assistant on a specific day. Generated answers vary between runs — our own case study says so in its limitations section. What we optimize is the probability of being cited across repeated samples, and that probability is exactly what the question set measures.

Ask for before-and-after results in your languages, the rules for counting mentions and citations, and the services tested. Our public case study documents our own site under the protocol above; your project starts with its own baseline. We record mentions and citations separately and test at least ChatGPT, Perplexity and Gemini in each language. We also edit pages for local readers, keep company names consistent across scripts where relevant, and work with independent sources that assistants use. Our USA guide identifies those platforms for the US market and provides the supporting data.

And the test we recommend running on any vendor, including us: ask for ten prompts where you are invisible today, the current answers, and the specific changes they would make. That test is literally our free audit — email hello@get-geo.ai and we will send yours back with the ten prompts attached.