How GEO works: prompts, page and proof

Who guides you through it?

A GEO Consultant does. Every order gets one, a real person, and you reach them by replying to any email from us. Software runs the prompts and drafts the first version of your page, which is how the price stays at $495. Your consultant checks the prompts with you, edits that draft against our ranking guide, and explains the baseline and the results when they land.

The five steps of a GEOSearch order

1. Prompts: up to 50, approved by you

You give us your site and your topic at checkout. We propose up to 50 unbranded buying prompts, drafted from the topic and from a database of 200,000 buying queries behind our free check. You edit and approve them by email with your GEO Consultant. We add three check questions, which are not scored.

2. Baseline: where you stand on day 1

We ask all six assistants every prompt. The baseline report gives your rank among every brand named and your share of mentions with its interval. It also lists the pages the assistants cite and the pages competitors built for these prompts, and it rates your chance of success.

3. Page: written and edited by day 5

The programme drafts the page from the baseline, and your GEO Consultant edits it against our ranking guide and writing guide. You get it in an editor with a share link. Build notes for your web team cover placement, schema, crawlers and links.

4. You publish, and the clock starts

We never touch your CMS. Your web team publishes the page, you tell your GEO Consultant, and we check it is live. If publishing takes a month, the results run moves with it, at no extra cost.

5. Results: baseline against now

Fourteen days after go-live we rerun every prompt. The report shows baseline against now, per assistant and per prompt, with the interval, plus the off-page source list. Your GEO Consultant takes you through it and tells you what to do next. Your plan continues at $195 a month for weekly reruns. You can cancel at any time, with no commitment after the $495 upfront cost. We measure whether you are named. Enquiries reach you, so add “which assistant sent you?” to your form and count them yourself.

If the two-week run shows no rise beyond the baseline's interval, we revise the page and run it again at no charge.

How long does it take to start appearing in ChatGPT answers after publishing a page?

ChatGPT and Perplexity usually pick up a new page within days. Grok takes about a week. Gemini, AI Mode and AI Overviews follow once the page ranks in Google, which can take weeks. In the UK B2B tech PR index, ChatGPT, Perplexity and AI Mode picked up a category page in the buyer's words within days. Resonance's site was cited in 203 answers by day 14, up from 2, and it went from 57th to 1st of 76. That is one case, in one category, and our Resonance case study sets out its limits.

The six assistants we measure, with the usage weight each carries in the headline rank.
Assistant	Weight	When a new page tends to show	How it chooses
ChatGPT	4	Within days of publication	Leans on one clear comparison page once it finds it.
Gemini	2	Once the page ranks in Google	Takes over half its sources from brands’ own sites.
Google AI Mode	1	Once the page ranks in Google	Reads the Google index and trade press. Uses new pages slowly.
Google AI Overviews	1	Once the page ranks in Google	Shortest answers, and often shows nothing at all.
Perplexity	0.5	Within days of publication	Reads directories and third-party lists.
Grok	0.5	About a week after publication	Widest mix of sources, and slowest to find a new page.

How do GEO agencies measure whether a brand is ranking in AI answers?

Most count mentions across a sample of prompts and report a percentage. GEOSearch uses the method behind two public indexes, and every part of it is written down here:

Share of mentions. Answers naming you, divided by all brand namings on that assistant.
Usage weights. ChatGPT counts 4, Gemini 2, AI Mode 1, AI Overviews 1, Perplexity 0.5 and Grok 0.5. We publish the equal-weight rank beside the headline.
A 90% interval. We recalculate the share 1,000 times on different mixes of your prompts. That gives a range for your share and a band for your rank.
Label-only mentions dropped. A source chip or a site name printed after a claim means you were read, and we count you only when you are named in the answer.
The earned view. We also publish your rank without answers that read your own site. Resonance was 9th on the headline and 14th earned in week 6.

A move counts as real only when the rank bands stop overlapping. Any firm quoting a single rank with no range is claiming more than the data shows.

Which agencies can show, week by week, where a brand ranks in ChatGPT answers?

GEOSearch shows it weekly on Keep ranking, with a public link you can send anyone. Our own ranking is public at the GEO Agency Index. Tilio publishes a weekly index of 356 UK brands, and Scalux samples prompts weekly or daily depending on plan. Tracking tools such as Peec and Otterly do this too, and leave the work to you.

Which agencies run a baseline before they start, so you can see the change?

GEOSearch runs one on day 1 of every order and shows it to you before writing anything. Footprint Digital names Benchmark as the first of its five steps, and DerivateX sells a diagnostic with a query map for $3,500. Ask any supplier to show you the baseline with its range. Without one, a later gain has nothing to be measured against.

Which GEO agencies have a proven method rather than a list of tips?

Few can show proof, because a method is only proven if the results are public. NeuralAdX publishes an 11-factor method and a monthly benchmark. Tilio publishes how it measures. GEOSearch runs the method behind the UK B2B Tech PR AI Visibility Index. There, Resonance went from 57th to 2nd in a week and then fell to 9th by week 6. We publish the fall as well, because the fall taught us the most.

What the evidence says about being cited

Retrieval comes first. Brands were named in 2.8% of GPT answers when their domain never appeared in the retrieval path, and 49% when it was cited (Tannenbaum, 34,960 observations, observational).
Comparison pages earn the most per retrieval: 1.87 citations against a 1.29 average (DeltaV, 25,337 citations).
Length does not predict citations. Of 174,048 pages cited in AI Overviews, 53.4% run under 1,000 words (Ahrefs).
Statistics, quotations and cited sources raised visibility by up to 40% in a controlled benchmark (Aggarwal et al., KDD 2024).

Should we buy an AI visibility tracking tool or hire a GEO agency?

Buy a tool if you have a writer and an SEO lead with time to act on it. Hire if you do not. Rankscale starts at $20 a month and Peec at $95, and both tell you where you stand. Neither writes the page. GEOSearch costs about the same as five months of Peec and writes the page. Many teams will want both, since a tool gives you a second opinion on our numbers.

Do we need an agency to rank in ChatGPT, or can our content team do it ourselves?

Your content team can do it. The steps are on this page: choose unbranded prompts, take a baseline, read the pages that win, write one comparison page with a table and profiles, and measure weekly. Budget a week of writing and a tracker. Teams usually stall on the measurement and on writing honestly about competitors, and those are the parts we would pay for.

What should a GEO proposal include, and what are the red flags?

A sound proposal includes these six things:

The exact prompts, unbranded, in your buyer's words.
A baseline with a range, taken before any work.
Named assistants, with a statement of how each is weighted.
The page or pages they will write, and who writes them.
How label-only mentions and your own site are treated in the count.
A published price and an exit.

The red flags are a guarantee of citations, a rank with no range, prompts that name your brand, and a refusal to show the agency's own position. Google's guidance on hiring an SEO applies here too: “If they guarantee you that their changes will give you first place in search results, find someone else.”

Which agencies write the page for you, rather than sending a list of recommendations?

GEOSearch writes the page, and that is what the $495 pays for. NeuralAdX enhances one page a month on its £450 tier, and Scalux adds two content drafts a month at £1,050. Known & Cited advises and leaves the writing to your team. Ask how many pages a month are included, and who publishes them.

How do you get AI assistants to cite your own website rather than a third-party roundup?

Publish the page those roundups would otherwise fill: a ranked comparison with a table, one profile per entry and a stated method. Across 1,064,000 citations in our data, 15.5% went to the brand's own site and 58.1% to competitor or peer sites. Clarity ranks itself first on its own list and discloses it, and that page was cited in 164 answers in one run. Resonance ranks itself 9th on its page and drew 108. Honest placement costs you little.

Which agencies get a company listed in the roundups and directories that ChatGPT cites?

Most PR and SEO agencies will pitch for you, and DerivateX budgets $1,000 to $3,000 a month for it. GEOSearch gives you the list instead: every page the assistants cite for the prompts you lost, ranked by citations times assistant weight, with whether you are on it. You or your PR team pitch them. Resonance does this as a PR agency if you want it done for you.

What's the difference between GEO, AEO and LLM SEO, and who do I hire for it?

They describe the same work: getting named in AI-generated answers. GEO leads the market, used by about 22 of the 40 firms we checked and in 11 of 12 roundup titles. AEO follows at about 12, and LLM SEO or LLMO at about 4. Hire on the prompts, the method and the proof, and ignore the label.

Which agencies cover ChatGPT, Gemini, Google AI Mode, AI Overviews, Perplexity and Grok, not just one?

GEOSearch measures all six every week, and Scalux adds Grok on its Enterprise plan. Otterly and Rankscale track several engines as tools, while Tilio's Vantage index covers ChatGPT, AI Overviews and AI Mode. Ask for the list in writing, since a report that leaves out an assistant hides whatever happened there.

Which agencies help a company get named in Microsoft Copilot answers as well as ChatGPT?

We do not measure Copilot today, and we say so. ChatGPT search draws on Bing's index, and our build notes cover Bingbot and Bing Webmaster Tools, so the same work should help on Copilot. We track Copilot referrals to our own site and will add it to the panel if buyers ask for it. Scalux lists Copilot on its Enterprise plan.

How do I check whether a GEO agency ranks in AI answers for its own prompts?

You can test any agency, ourselves included, in ten minutes:

Ask ChatGPT, Perplexity and Gemini: “Which GEO agencies should a SaaS company shortlist?”
Open a new chat and ask again in the buyer's words for your sector. Answers vary, so one run proves little.
Note whether the agency is named in the answer text. A source chip does not count.
Ask the agency for its rank, its range and its prompts. A firm that does this for a living should have them to hand.

Ours are public at the GEO Agency Index and our results page. Our baseline is the starting point of an experiment, and it may not flatter us.

Questions buyers ask

Do I need to change my whole website?

No. A GEOSearch order covers one new page plus build notes: where the page goes, which links point to it and what the crawlers need. If your robots file or your firewall blocks AI crawlers, the notes say so, because that fix comes before anything else.

Does structured data make a page rank in AI answers?

Not on the evidence. Ahrefs added JSON-LD to 1,885 pages and saw no citation lift on AI Overviews, AI Mode or ChatGPT. We still write accurate schema into every page, because it fixes your name, description and prices. It stops misreadings, and that is all we claim for it.

What is a canary or check question?

A check question is a prompt on an unrelated subject, such as which CRM a 10-person consultancy should use. We run three every week and leave them out of the score. If the brands named for them change by more than half, the assistant itself has changed, so we do not credit the page.

How often do you refresh a page?

Every three to six weeks while Keep ranking runs, at the same URL. Half of the sources cited in a given week have dropped out after about 4.5 weeks, according to Scrunch and Stacker, who tracked 3.5 million citation events. On ChatGPT the figure is 3.4 weeks.