How to choose a GEO agency: the questions that expose a fake one

Updated October 2026 7 min read

Should you trust a buyer's guide written by a GEO agency?

Only as far as it gives you checks you can run without us. Redaccel sells GEO retainers, so we have an interest in you hiring someone. What follows is the checklist we would want a buyer to use on us, including the parts that argue for hiring nobody. Every figure is linked and dated, and where nobody has measured something, we say so.

One disclosure up front. In our own 30-day tracking of 28 unbranded buyer questions across ChatGPT, Gemini and Google AI Overviews (25 August to 24 September 2026, 2,594 answers), Redaccel was named in none of them^[5]. The engines mostly named four other agencies, drawn from a small set of list pages. You can read the full method on our GEO agencies compared page.

What is the difference between a GEO agency and an SEO agency?

An SEO agency works to move your page up a ranked list; a GEO agency works to get your brand named and cited inside an AI answer, which is a paragraph with no positions. If you need the background, start with what GEO is and how it relates to GEO versus SEO.

The overlap is real. Google's own guide says SEO best practices "continue to be relevant" because its generative AI features "are rooted in our core Search ranking and quality systems"^[3]. But ranking well no longer means being cited. In Ahrefs' March 2026 analysis of 863,000 keyword results and 4 million AI Overview URLs, 37.9% of cited URLs also ranked in the top 10, down from about 76% in its July 2025 study^[2]. Roughly 31% of citations came from pages ranked 11 to 100, and another 31% from pages beyond the top 100^[2].

That gap is the case for GEO. It is not, by itself, a case for a new agency.

When should you not hire a GEO agency at all?

Skip the agency if your existing SEO program is healthy and nobody has measured a gap in AI answers yet. Measure first. In our experience, these are the situations where hiring anyone is premature:

• You have no baseline. If you don't know how often engines name you today, you cannot tell whether an agency helped. A one-off audit costs less than three months of retainer.
• Your buyers don't ask AI. Some categories, such as local trades or very small B2B niches, may see few AI-assisted searches. Check your analytics for AI referrals before you assume otherwise.
• Your site has basic problems. If pages are not indexed or load badly, fix that first. Google says AI features rely on the same Search systems^[3], so broken foundations cap any GEO work.
• You already show up. If engines already name you in most buyer prompts, an existing SEO and PR program may be enough. Keep monitoring, and spend on a retainer only when the numbers slip.

What does a real AI visibility audit show?

A real audit names the engines, the prompt count and the time window, and lists which pages the engines cite today. If any of those four is missing, you are looking at a sales deck.

The reason is variance. SparkToro and Gumshoe had 600 volunteers run 12 prompts through ChatGPT, Claude and Google's AI a combined 2,961 times in November and December 2025^[1]. They found a less than 1 in 100 chance that ChatGPT or Google's AI would return the same list of brands in any two responses, and roughly 1 in 1,000 for the same order^[1]. Their conclusion was that visibility percentage "across dozens to hundreds of prompts run multiple times is a reasonable metric", while rank position is not^[1].

So a credible audit looks like this:

• Engines: each one named, for example ChatGPT, Gemini, Perplexity, Google AI Overviews and AI Mode, because they cite different sources.
• Prompts: dozens at least, written the way buyers ask, split by intent (discovery, comparison, pricing, problem first), with the full list handed over.
• Time window: repeated runs over days or weeks, with dates stated. Our own comparison used 28 prompts, run daily for 30 days^[5].
• Cited pages: the specific URLs each engine retrieves and cites for your category today, yours and other people's. This is the map the work is planned from.
• Share of voice: how often you are named versus each competitor, as a percentage of answers.

Read more

Ask whether the audit separates mentions from citations. A brand can be named without its site being linked, and a site can be cited without the brand being recommended. Both matter, but they move for different reasons, and an agency that blends them into one score is hiding which one it can actually change.

Which questions expose a fake GEO agency on a sales call?

Ask for specifics you can verify later, and watch for answers that drift back to rankings, markup or guarantees. These are the questions we would ask.

Push hardest on the last row, because it catches the easiest fake to sell: an SEO package relabelled as GEO. It is not dishonest to sell SEO. It is dishonest to sell it under a new name at a new price.

What does a real GEO deliverable look like?

A real deliverable is mostly content and sources, not markup. Schema and llms.txt are a small part of the work, and Google says so directly: "Structured data isn't required for generative AI search, and there's no special schema.org markup you need to add"^[3]. The same guide says you don't need new "AI text files" to appear in Google Search, and don't need to break content into tiny pieces for AI^[3]. Other engines publish less guidance, so treat markup as hygiene, not strategy.

What does seem to matter is the content itself. In the original GEO research paper (Princeton and others, KDD 2024), methods such as adding citations, quotations and statistics to a page lifted visibility in generative answers by up to 40%, while packing in keywords did little^[4]. That was a lab benchmark from 2023 to 2024, so read it as direction, not a forecast.

In practice, a monthly deliverable from a serious agency should include:

• A share of voice report on the agreed prompt set, per engine, with the cited sources behind each answer.
• Owned pages rewritten or created so they answer buyer questions directly, with sourced numbers an engine can quote.
• Work on the third-party pages engines already cite: lists, reviews, forums, trade press. In our tracking, a handful of list pages and Reddit threads drove most agency mentions^[5].
• A plain statement of what moved, what didn't, and what changes next month.

What does a GEO agency cost, and how do you measure ROI?

Our own band is EUR 2,000 to 5,000+ a month, with a three-month minimum, scoped after an audit; the plans are on our pricing page. That is our price, not a market survey. We have not found a published, sourced study of GEO agency pricing, so we won't print a market range. Quotes far below a retainer usually mean monitoring only; quotes far above should come with a named team and named outlets.

Measure ROI the same way you would audit the agency:

1. Baseline before signing. Share of voice per engine on a fixed prompt set, with dates.
2. Same prompts, re-measured inside a quarter. Change the prompt set and you can no longer compare.
3. Read movement against variance. Because single runs disagree so often^[1], compare averages over a window. A jump in one screenshot is noise until repeated runs confirm it.
4. Tie it to the pipeline where you can. AI referrals in analytics and "where did you hear about us" answers are imperfect, but they beat no link at all.

If share of voice has not moved after a quarter, the agency should say so plainly and propose a change or an exit. That conversation is a fair test of whether you hired well.

Frequently asked

Sources

1. SparkToro (Rand Fishkin, with Gumshoe.ai), "New Research: AIs are highly inconsistent when recommending brands or products", January 2026. 600 volunteers, 12 prompts, 2,961 runs, November to December 2025.
2. Ahrefs (Louise Linehan), "Update: 38% of AI Overview Citations Pull From The Top 10", March 2026. 863,000 keyword SERPs, 4 million AI Overview URLs.
3. Google Search Central, "Optimizing your website for generative AI features on Google Search", last updated July 2026.
4. Aggarwal, Murahari, Rajpurohit, Kalyan, Narasimhan and Deshpande, "GEO: Generative Engine Optimization", KDD 2024 (arXiv v3, June 2024).
5. Redaccel, "Which GEO agencies do AI engines actually recommend? 30 days of citation data", September 2026. 28 prompts, ChatGPT, Gemini and Google AI Overviews, 25 August to 24 September 2026, 2,594 answers.