Generative Engine Optimization (GEO) agency: PR-led visibility in AI answers
What generative engine optimization is
Generative engine optimization (GEO) — also called AI search optimization, LLM SEO or answer engine optimization (AEO) — is the work of preparing a brand and its content so that large language models mention it in their answers and describe it accurately. The names differ; the job is the same.
It is not a replacement for SEO. Search still matters, and tools like Gemini lean heavily on what ranks in Google. But a growing share of buying research now ends inside an AI answer, before anyone clicks a link. If the answer does not name you, the click never happens.
Where AI models get their knowledge about companies
This is the question the whole discipline stands on, and the one with measured answers. The largest study so far analysed more than 25 million links cited by ChatGPT, Claude and Gemini (Muck Rack, What Is AI Reading?, May 2026):
- 84% of cited sources are earned media — articles, research, public sources and independent sites the brand did not pay for.
- 0.3% are paid content. You cannot buy a place in a model’s answer, and there is no ad console for it.
- 27% of citations are journalism; for questions about industry trends, the share of answers citing news media reaches 46%.
- More than half of cited texts are under a year old, and citations drop noticeably after six months. Freshness is a condition, not a bonus.
The conclusion is uncomfortable for ad budgets and good news for companies with a real track record: in AI answers, the brand others write about wins — not the brand that talks most about itself. That is why most GEO work is public relations work, not on-page tweaking.
Why a PR agency, not only an SEO shop
Most GEO providers come from technical SEO. They fix schema, rewrite pages and publish more content on your own site. That helps — but your own site is a minority of the signal models use. If 84% of what they cite is earned media, the lever is getting you into the sources they trust: trade press, national media, research reports and independent reviews.
That is what we have done for three decades. We combine it with the technical side — structured data, clear entity information, pages written to be quoted — so both halves of the signal point the same way.
Every model answers differently — and that changes the plan
Treating “AI” as one thing is the most common planning mistake. The four models buyers use most behave measurably differently (same Muck Rack study):
- ChatGPT attaches sources to 96% of answers — the most citation-heavy, so a strong publication moves it most directly.
- Gemini cites sources in 82% of answers and is closely tied to what it sees in Google results.
- Claude cites in 55% of answers; it cites less often and more selectively, so being in its sources weighs more.
- Perplexity always shows sources and is used most directly as a search engine.
A plan that lifts visibility in one model will not automatically work in the other three. We always measure all four.
What we measure
Six things, each with a number, so progress is a fact rather than an impression:
- Presence — in how many answers to how many buying questions your brand appears at all.
- Position in the answer — named first, or last in a list of five vendors.
- Accuracy — whether what the model says about your services and facts is true.
- Sources — which sites the model draws on for your category, and whether you are in any of them.
- Share against competitors — who else is named for the same questions, and how often.
- Freshness — how old the cited material about your brand is.
A monthly measurement on the same fixed question set is what separates GEO work from talking about GEO.
How an engagement runs
- Baseline audit. We define the buying questions your customers actually ask, run them across four models and record presence, position, accuracy and sources.
- Source map. We list the publications and sites the models cite for your category and mark where you are missing.
- Plan. Earned media targets, expert content, corrections to inaccurate descriptions and technical fixes on your site — with an owner and a date for each.
- Execution. Media relations, research and commentary placements, entity and structured-data work.
- Monthly re-measurement. Same questions, same models, compared with the baseline.
Four mistakes that make GEO fail
- Treating it as SEO. On-site changes help, but your site is the smaller part of the signal. A company that only changes its own pages measures its own effort, not its visibility.
- One-off checks. Model answers change week to week. A single screenshot is an anecdote, not a measurement.
- Sponsored content as a shortcut. Paid content is 0.3% of citations. Real spend, near-zero effect on AI visibility.
- Stopping after three months. Citations fade after about six months. AI visibility is a state you maintain, not a project with an end date.
Who we work with
Companies that sell in Poland, or want to — including international brands entering the Polish market. Models answer differently in Polish and in English, and cite different sources in each language. We work in both, with local media relationships that an overseas SEO provider does not have.
What drives the budget
Not the number of articles. Four variables: how many buying questions we measure, how many markets and languages, the brand’s starting point, and the pace at which media publish — which cannot be bought faster. We quote a monthly range after the baseline audit, because before it any number would be a guess.
Frequently asked questions
What is the difference between GEO and SEO?
SEO aims to rank your pages in search results. GEO aims to get your brand named and correctly described inside AI-generated answers. They overlap — Gemini and Perplexity use search results — but GEO depends much more on what independent sources say about you.
Can you guarantee that ChatGPT will recommend my company?
No, and nobody honestly can. Models are not controlled by any vendor. What we can do is measure your starting point, change the inputs models rely on, and show the movement month by month on a fixed question set.
How long does it take to see results?
Technical fixes and corrections to your own content can show within weeks. Earned media builds over months, at the pace publications appear. We report monthly so you see the direction early.
Is GEO the same as AEO?
In marketing, yes — both describe optimizing for AI-generated answers. Note that in Poland “AEO” most often means the customs status Authorised Economic Operator, a different audience entirely.
Do you work outside Poland?
Our core market is Poland, and we work in Polish and English. For brands selling into Poland from abroad, local media access is the point.
How do you measure AI visibility?
We run a fixed set of buying questions across ChatGPT, Gemini, Claude and Perplexity every month and record presence, position, accuracy, sources, share against competitors and freshness of citations.
Who is behind this
Commplace® has worked in public relations since 1996. We treat AI visibility as an extension of what we have always done: earning coverage in sources people trust. What changed is that machines now read them too.
Author: Sebastian Kopiej, CEO of Commplace®. In public relations since 1996. Written with the help of AI tools and editorially verified by the author. Data current as of 22 September 2026.