- OpenAI states that placement in ChatGPT search is not guaranteed, so a guaranteed ranking is a red flag.
- AI answers vary from run to run: in a SparkToro study, tools almost never returned the same brand list twice across 100 runs.
- Many “best GEO agency” lists are published by agencies themselves; in our measurement, even an AI assistant warned about it.
- A credible case shows both measurements and the objectives missed, not only the wins.
What questions should I ask a GEO agency before signing?
Ask eight questions that force concrete answers about measurement, production and proof. A serious agency answers them in writing before you sign; vague answers to any of them are a reason to keep looking.
| Question to ask | A good answer | Red flag |
|---|---|---|
| Which buying questions will you measure, and who approves them? | A list of your real buyer questions, agreed with you before the baseline | Generic keywords, or a list you never see |
| How many runs per question and engine? | Several runs, engine by engine, with the number stated | One run, or screenshots |
| What noise threshold do you use, and how was it set? | A threshold measured by running the same protocol twice, set before any comparison | No threshold, or one chosen after the results |
| Who writes, who checks facts, and who publishes? | Named roles, on your site and on third-party sources | “We provide briefs” when you need execution |
| Do I get the raw answers? | Yes, every answer behind every number | Only a score or a chart |
| Can you guarantee results? | No; commitments on means and on the method | A guaranteed position or mention |
| What commitment, and why that length? | A duration tied to pickup time and re-measurement | A long lock-in with no measurement in between |
| Can you show a before/after? | Both measurements, same protocol, including objectives missed | Only the “after”, or only wins |
Then check whether the agency is described accurately in AI answers about its own field, and whether it measures itself with the protocol it sells you.
Can a GEO agency guarantee results in ChatGPT?
No agency can guarantee it honestly. OpenAI states that placement in ChatGPT search is not guaranteed, and the same question returns different brands from one run to the next [1][2].
ChatGPT search rewrites a question into targeted searches, pulls results from third-party search providers and ranks them on several factors [1]. An agency controls none of those systems. What it does control: the questions measured, the content it produces and publishes, the protocol, and the honesty of the report. A contract should commit to those means, and to objectives measured with the same protocol, reported even when they are missed.
What methodology should a GEO agency use to measure results?
Repeated measurement on a fixed question set, with “named”, “cited” and “absent” counted separately and every change compared with a noise threshold set in advance. Anything less turns normal variation into fake progress.
In a SparkToro study, AI tools almost never returned the same list of brands twice across 100 runs of the same prompt [2]. That is why one answer is an anecdote. Ask for:
- A frozen question set, with the questions that already contain your brand name set apart, since they inflate the score.
- Several runs per question and engine, reported engine by engine: ChatGPT and Perplexity rarely move together.
- A noise threshold measured by running the full protocol twice before any comparison.
- A validated competitor list if the agency reports share of voice, so the number stays comparable from month to month.
- The raw answers, so you or a third party can recount.
Are "top GEO agency" rankings trustworthy?
Treat them with care. Many are published by agencies that rank themselves first, and AI engines still cite them, so check who publishes each list, whether the method is public and whether the data can be checked.
In our measurement of 65 US buyer questions about GEO (1,300 answers + 325 Google searches), agencies’ own sites and third-party “best of” lists were among the most cited sources on “which agency” questions [6]. In one answer, Claude itself warned the user that many “best agency” lists come from the agencies being ranked. ZivRank publishes rankings too, with its method and sources: ask it the same three questions. A list built on repeated measurement in AI engines deserves more weight than an editorial one.
Can you manipulate ChatGPT answers? What are the risks?
Some agencies try, with hidden text, fake reviews, fake forum accounts or pages produced at scale. The risks are legal, technical and commercial, and they land on your brand, not on the agency.
- Google’s spam policies target scaled content abuse and hidden text [3].
- Reddit’s rules prohibit manipulation and spam [4].
- In the US, the FTC rule on consumer reviews and testimonials bans fake reviews [5].
Ask the agency to describe, in writing, how it places content off your site and under which accounts. If the answer involves accounts you do not control or reviews you did not request, walk away.
Should a GEO agency show up in AI answers itself?
It is a fair test, read with nuance: a young or specialized agency may not yet be named on broad questions. What matters is that it measures itself with the protocol it sells and tells you the result plainly.
Ask the engines your own version of “which GEO agency for my sector” several times and see who is named and from which sources. Then ask the agency for its own baseline. Ask how the agency measures itself, and with which protocol.
How does ZivRank answer these questions?
In writing, before you sign, with the figures below taken from our published offers and protocol. Check each answer against the pages linked here.
- Questions and runs: 50 buying questions agreed with you, each asked 5 times on ChatGPT, Claude, Gemini, Perplexity and Google AI Overviews: 1,000 answers + 250 Google searches per measurement.
- Noise threshold: ±15%, set before any comparison by running the full protocol twice the same night.
- Production: our team writes, fact-checks and publishes, on your site and off it. You receive the raw answers.
- Commitment and price: 6 months, because pickup takes 3 to 8 weeks and needs re-measurement. Programs $2,300 to $10,000 per month; baseline audit $2,000, billed at the start and included in your 6-month program.
- Before/after: our published case, a US B2B SaaS vendor (ITSM), was measured twice, 44 days apart: answers naming the brand on Perplexity went from 21 to 50, while ChatGPT stayed at 17. The case page lists the objectives that were missed, too.
- Self-measurement: we measure our own market with the same method (repeated runs, every cited domain logged); the figures on this page come from it.
The full protocol is on the methodology page and the case on the US B2B SaaS vendor case study.
Frequently asked questions
Should I hire a freelance GEO consultant or an agency?
Hire a freelancer for a bounded job, such as an audit, a strategy or a set of pages your team will publish. Hire an agency when you need measurement, writing, publishing and off-site distribution running at the same time, month after month. Apply the same checks to both: a baseline on your questions, repeated runs and a before/after.
How much does a GEO agency cost per month?
Published prices range from a few hundred to several thousand dollars a month, driven by the number of questions tracked and the volume of content produced. At ZivRank, tracking (waitlist) starts at $390 per month and programs run $2,300 to $10,000 per month, over 6 months. Compare what each price produces, not only the monthly figure.
What minimum commitment is reasonable with a GEO agency?
Long enough to publish, wait for pickup and re-measure at least once. With pickup observed at 3 to 8 weeks after publication, three months leaves little room for a second cycle. Six months is a common minimum. Twelve months makes sense when the agency offers a real discount for it and reports results along the way.
Are there GEO case studies with measured B2B results?
A handful are public. Read them for the method: both measurements, the same question set, the number of runs, engine-by-engine results, and the objectives that were missed. A case that shows only an “after” figure, or a single engine, says little about what the agency will do for you.
Sources
Every figure in this article comes from the sources below, checked on November 9, 2026. Bracketed numbers in the text refer to this list.
- [1] OpenAI Help Center, Searching the web with ChatGPT, checked October 11, 2026.
- [2] SparkToro (Rand Fishkin), NEW Research: AIs are highly inconsistent when recommending brands or products, January 27, 2026.
- [3] Google Search Central, Spam policies for Google web search, checked November 9, 2026.
- [4] Reddit, Reddit Rules, checked November 9, 2026.
- [5] Federal Trade Commission, Trade Regulation Rule on the Use of Consumer Reviews and Testimonials, posted August 21, 2024.
- [6] ZivRank, US measurement: 65 buyer questions about GEO, ChatGPT, Claude, Gemini and Perplexity × 5 runs (1,300 answers) + 325 Google searches, November 8, 2026.