How to Vet a GEO Agency: 7 Questions for a "ChatGPT Visibility in 3 Months" Promise

Key takeaway: A GEO agency proposal that promises "ChatGPT visibility in 3 months" can only be verified if it states what counts as visibility, which questions are used, which AI services are measured and how many runs decide the result. AI answers change even when the same question is repeated, and when RanketAI measured 12 Korean B2B SaaS companies twice on the same day, individual scores moved by up to 17 points. Without measurement conditions, you cannot tell whether a result is progress or measurement noise. This guide covers seven questions to ask before signing, proposal phrases to double-check and a baseline you can measure yourself beforehand.

---

Three-line summary

• A visibility guarantee can be verified only when it states the definition of visibility, the questions, the AI services and the number of runs.
• Branded questions, single runs and rank guarantees are conditions that make results look better than they are.
• If you measure a baseline yourself with the same questions before signing, you can check later reports under the same conditions.

Why a visibility guarantee cannot be taken at face value

AI answers change even for the same question, so a guarantee cannot be verified unless repeated measurement under the same conditions shows the promised change is larger than measurement noise.

In a study by SparkToro and Gumshoe that repeated the same prompts, the chance that ChatGPT or Google's AI search returned the same list of brands was under 1 in 100, and the same list in the same order appeared less than 0.1% of the time (Search Engine Journal, 2026-01-30). When RanketAI measured 12 Korean B2B SaaS companies twice on the same day, individual scores moved by up to 17 points (median 6) and 2 of 12 changed grade (August 2026 benchmark). In the October issue, after a measurement-engine revision, movement for well-known brands fell to at most 6 points, but the report notes that brands near a grade boundary can move more.

Search engine optimization has long had the same principle. Google's guidance on hiring an SEO says:

"No one can guarantee a #1 ranking on Google." — Google Search Central, Do you need an SEO? (updated 2026-06-05)

The same page advises finding someone else if an SEO guarantees first place, and to beware of those claiming a "special relationship" with Google or a "priority submit." In AI search, where answers change every time, this principle should be applied even more strictly.

Seven questions to vet the proposal

A proposal that gives concrete answers to these seven questions can be verified; if the answers are vague, get them in writing before you sign.

1. What counts as "visibility"?

Visibility can mean a mention (your brand name in the answer text), a citation (your site in the answer's source list) or a position in the recommendation list. The same result reads very differently depending on the definition. In RanketAI's October measurement, ChatGPT, Gemini and Perplexity all mentioned 83% of brands, while own-site citation ranged from 33% to 83%. Position is hard to guarantee, since the same order reappears less than 0.1% of the time.

2. Which questions will you measure?

Get the list of measurement questions before signing. A question containing your brand name ("Tell me about our company") almost always gets your brand mentioned by an AI with web search, so it shows no visibility gain (the branded-query trap). Confirm that the questions are unbranded ones your customers actually ask, and that the same list is used before and after.

3. Which AI services, and do you separate answers with and without web search?

Confirm whether "ChatGPT visibility" means ChatGPT only or also Gemini and Perplexity. Results differ by service. When RanketAI measured with web search enabled, every ChatGPT and Perplexity answer ran a search, but 62% of Gemini answers came without a search and had no sources (October measurement analysis). No amount of work can get your site cited in an answer with no sources, so check whether the report separates the two.

4. How many runs of each question decide the result?

Judging "visible" from a single run reports measurement noise as progress. Check that results are reported as "mentioned 2 out of 3 times" rather than a single outcome, and that same-day repeated runs show the range of variation. How many repetitions you need is covered in our measurement statistics post.

5. Will you measure and share a baseline under the same conditions before the contract?

A claim of improvement can only be checked against a recorded starting point. Get a written baseline measured with the same questions, AI services and conditions before the contract, and confirm later reports use exactly those conditions.

6. What exactly will you do yourselves?

Scope varies by agency. Check item by item whether they only deliver a diagnostic report, fix site structure and content themselves, publish content on external media or communities, and include re-measurement reports. If external publishing is included, ask where, whether it is paid and whether it is labeled as advertising. Paid mentions can be judged as violating search engines' spam policies (the paid-mention gray market), and in one experiment self-promotional comparison posts were cited without mentioning the brand 43% of the time (self-promotion listicle analysis). Types of GEO services in Korea are summarized in our Korean GEO tools comparison.

7. What happens if the target is missed?

Write into the contract what happens when the guaranteed level is not reached — a refund, an extension — and whose measurement counts if results are disputed. Also agree in advance whether changes within the measurement-noise range count as progress.

Proposal phrases to double-check

These phrases do not make a proposal wrong by themselves, but you need the measurement conditions and evidence in writing to check the result.

Proposal phrase	What to check
"Guaranteed #1 ChatGPT recommendation"	The same order reappears less than 0.1% of the time. Check what counts as #1 and how many runs out of how many
"Confirmed AI visibility for brand searches"	Branded questions almost always get a mention. Ask for results on unbranded questions
"Dedicated AI Overviews optimization"	Google says there are no additional requirements or special optimizations for AI Overviews or AI Mode (Google Search Central). Check how the work differs from regular SEO
"N press articles distributed"	Check whether placement is paid, labeled as advertising and on which outlets
"Partnered with AI platforms for priority inclusion"	Google warns against SEOs claiming a special relationship with the search engine. Verify any partnership through official sources
"Results within a month"	Scores move even on same-day re-measurement. Check whether repeated runs under the same conditions show a change larger than that

A baseline to measure yourself before signing

If you record your brand's current state with the same questions before the contract, you can check the reports you receive afterward under the same conditions.

1. Choose 3–5 unbranded questions your customers actually ask.
2. Check each question three times in ChatGPT, Gemini and Perplexity, in a fresh chat each time.
3. For each answer, record whether your brand is mentioned, whether there is a source list, whether your site is among the sources, and which competing brands appear instead.
4. Ask the agency to include the same questions in its measurement.
5. After the contract, re-measure at the same intervals under the same conditions and compare with the report.

The detailed manual method is in How to Check If ChatGPT Mentions Your Brand: 5 Steps. To automate the measurement, you can use AI visibility tools such as RanketAI. RanketAI's AI brand visibility analysis measures mentions and source citations in ChatGPT, Gemini and Perplexity answers, and competitor comparison measures competing brands side by side with the same questions. You can try brand visibility analysis three times for free after signing in.

FAQ

Related reading

• How to Check If ChatGPT Mentions Your Brand: 5 Steps (and When Only Competitors Show Up) — how to measure your own baseline before signing.
• Monthly Change Was Smaller Than Same-Day Measurement Noise (August 2026 Benchmark) — how to separate "it went up" from measurement noise.
• Buying AI Citations: The Paid Brand Mention Gray Market and the 2026 Spam Crackdown — when external publishing becomes risky.
• Korean GEO Tools Comparison — 5 Categories of AI Visibility Diagnostics — GEO services in Korea by type.