Our GEO methodology
SEO Back Office is a GEO agency. This page writes out the method we run on every engagement, what we report, where the numbers come from and what they can't tell you.
How does a GEO engagement work?
A GEO engagement runs in eight stages, from building the query set your buyers use through to reporting AI mentions next to booked calls. The early stages measure what’s there. The later ones change it. Then the loop repeats every month so the numbers stay comparable.
Nothing here is proprietary. We publish it because a client should be able to check the work, and because an agency that won’t say what it does is usually doing less than it charges for. What we sell is running this properly, every month, on your category. A GEO agency that can’t show you a page like this one is asking for trust it hasn’t earned.
The eight stages
- Query universe. Forty to eighty questions built from what your buyers type into an AI, not from a keyword tool, and you approve the set before anything runs.
- Baseline. Every query runs across ChatGPT, Perplexity, Gemini, Claude and Google’s AI Overviews and AI Mode, repeated over two to three weeks, with four fields logged each time. Whether you were named, where you appeared, which pages got cited, and how you were described.
- Source and citation mapping. The cited URLs get ranked by how often they appear across your query set. That ranking is the outreach list, and it’s usually shorter and stranger than anyone expects.
- Technical and entity fixes. Crawler access confirmed for GPTBot, OAI-SearchBot, PerplexityBot, ClaudeBot and Google-Extended, the important content checked so it doesn’t arrive only after a script runs, schema corrected, and one description of the company agreed and pushed everywhere your brand is listed.
- Owned content. Answer-first pages on your site that say the thing in the first line under the heading, with claims specific enough for a model to lift, plus the comparison and best-of pages your buyers already search for.
- Third-party authority. Outreach to the publishers on the citation list, one at a time, to get you added where you’re missing and corrected where you’re wrong. Every ask, every reply and every refusal gets logged.
- Repeated prompt testing. The same query set, the same fields, every month. Comparability is the whole value, so the set changes only when you and we agree to change it, and the report says when it did.
- Revenue attribution. The AI numbers get reported next to branded search volume, qualified visits and booked calls, because a mention that moves none of those is a vanity metric with better marketing.
What gets reported and how
The monthly report carries mention rate, citation rate, competitor mention rate, the cited URLs and the cited domains, split by platform. All of it comes from the same query set run on the same schedule, so a rise means something happened rather than that we changed the questions.
Underneath the numbers sit the working documents. The outreach tracker shows every publisher contacted, what they asked for and what landed. The correction tracker shows what got fixed and where. Every citation in the report is a link you can open and check yourself.
We don’t publish a proprietary visibility score. A single number would make the report prettier and hide what’s worth reading, which is the list of pages behind the answers and whether your brand is on them.
Where the data comes from
The prompt runs are ours. We run the query set against the live consumer products and keep the raw output, so any number in the report traces back to the run that produced it.
Everything else comes from places you already have access to. Search Console for branded queries and impressions, your analytics for referrals from the AI products that pass a referrer, your CRM for the calls. Crawler behaviour comes out of server logs when you can give us them, and out of robots.txt plus a fetch test when you can’t.
Google’s published guidance shapes how we read all of it. The 2026 note on optimizing for generative AI features says those features are rooted in core Search ranking systems, which is why ranking stays in the plan, and the 2025 documentation said there are no extra requirements for appearing in AI Overviews beyond Search fundamentals, which is why we don’t sell AI-specific formatting.
What this method can’t tell you
It can’t tell you what any one person saw. Answers vary between accounts, locations and moments, so the baseline is a rate across many runs rather than a fact about a single answer.
It can’t separate your citation work from the rest of your marketing with any rigour. If a brand gets mentioned more after a funding round, a product launch and three months of outreach, the log shows the mentions and the honest read is that all of it contributed.
And it can’t make a publisher say yes. Some decline, some want payment, some have a policy against listing vendors at all, and we report the refusals alongside the placements. The GEO services page breaks each stage into something you can buy, and the AI visibility audit is where the first two stages get delivered.
Run it on your category.
Send us your category and your three closest competitors. We’ll run the first round of queries before the call, so the conversation starts from the answers rather than from a pitch deck.
Request an AI visibility audit No inflated promises. A clear baseline and a practical roadmap.