Most AI answers name no source at all.They still describe your category. And you.
Generative engine optimization for companies that have worked out their buyers are arriving already informed by something nobody in the building wrote.
You can be invisible in AI answers and never see it happen. No ranking drops, no traffic alert. Just fewer of the right conversations, and a competitor whose name the model learned first.
Three questions in a stack, and GEO is the last one
SEO asks: can they find you?
Crawling, indexing, rendering. The foundation, and the layer everything else stands on. Measured in impressions, position and clicks.
AEO asks: will they quote you?
Structure, schema and entity clarity, so an assistant can lift an answer from your page and credit you for it. Measured in citations and brand mentions.
GEO asks: what do they say when they cite nobody?
Most AI answers name no source at all. What the model believes about your category, your brand and your competitors is what the buyer hears. Measured in share of voice and sentiment, not links.
This is why the three are sold and measured separately. Search is the foundation, answer engine optimization is the middle layer, and GEO is the part almost nobody measures because it has no link to count.
It reaches further than most people expect. Local answers increasingly arrive without a map, and product recommendations arrive without a comparison page. In both cases the model is describing you from whatever it absorbed.
If you are looking for a GEO agency, start here
You will find two kinds of result for that search, and they do very different jobs.
What a GEO platform does
Tracks prompts. Counts citations. Draws a chart. Tells you the number went down.
Useful, and we run these tools every week. None of them changes a single thing on your website.
What a GEO agency does
Reads the chart, works out which of nine possible causes is the real one, fixes it in the right order, and proves the fix reached the crawlers that matter.
We are the second thing. Small, senior, and we do the work.
We are a small agency and we say so plainly. No account manager, nobody junior learning on your account, and a network of specialists when the work needs them. The person who wins the work does the work.
What measured AI visibility looks like
One client, tracked across the questions its buyers actually ask. It is the second most used source in its category, ahead of a national retailer, a major review publisher and two government agencies. Only a forum is used more.
Instrument and window: Searchable, 30 days to September 24, 2026, across ChatGPT, Perplexity and Google AI Overviews. Rank is by share of answers that draw on the source, not by raw citation count.
Distinct pages cited in AI answers, same category, 30 days
Now read the first and last numbers together. Second most used source, named in 5.8% of the answers it feeds. Cited constantly, credited almost never. That gap is what this work closes, and a tool reporting citations alone would have called it a win.
The problem is usually structural, not editorial
A real diagnosis, on a live account
The client’s entire organization schema was invisible to the crawlers it was built for.
It had been delivered through a tag manager, so it existed only after JavaScript ran. Google renders JavaScript. GPTBot, ClaudeBot, PerplexityBot and CCBot generally do not.
The only statement of who that company was could not be read by any of the engines the program existed to reach. The fix was serving it from the server, not writing anything new. Verified afterwards by direct fetch and by a headless render.
This is the shape of most GEO work. Not more words, but making what you already have reachable, parseable and attributable to you. Unglamorous, and it is where the movement comes from.
Why most GEO programs produce nothing
Buying the tool and stopping
A dashboard tells you that visibility fell. It cannot tell you which of nine causes did it, and it will not fix any of them.
Publishing more, hoping harder
Volume is the most expensive way to fail at this. If the model cannot parse or attribute what you already have, the next fifty articles inherit the same problem.
Ignoring what other people say about you
Models learn your category from sources you do not own. Forums, review sites and trade publications frequently outrank your own site as a description of you.
Working out of sequence
Entity work before indexation, or content before rendering, produces no compounding. Each layer depends on the one beneath it.
The real cost is not the wasted budget. It is the compounding advantage a structurally clearer competitor builds while your fixes stay disconnected.
And our own scoreboard
Our own numbers, because you should ask
We track 61 non-branded prompts about AI search, GEO and digital advisory. We are mentioned in none of them.
Ask an assistant for a GEO consultant for mid-market B2B companies and it will cite seven sources. None of them is us. Our own visibility score is 7 out of 100.
We are telling you this on the page that sells the service. We are early in our own program, we recorded the baseline before starting it, and you will be able to watch whether it moves. An agency that will not show you its own scoreboard is showing you someone else’s.
How we work
Diagnose
What models currently say about you, which sources they trust instead of you, and which structural gaps explain it. Bounded, paid, and it ends with something you could act on without us.
Sequence
A remediation order, because these fixes only compound in dependency order. Ranked by business impact rather than by how easy they are to do.
Execute and enable
We do the work with your team rather than behind a portal. The capability stays with you, which is the only version of this that survives us leaving.
You can stop after the diagnostic, and some clients do. That is a normal outcome rather than a failed one.
What you will see, and when
The prompt set and the baseline are recorded. Publish first and the baseline is gone permanently.
Structural signals: indexation, rendering, schema validity, entity coverage. These move first because we control them.
Share of voice, citations and how the model describes you. Slower, and a 28-day read alone would mislead you.
What the instruments cannot do, stated plainly. AI answers are non-deterministic, so a single query is a sample of one and we re-run any absence claim before it reaches you. Citation exports refresh only partially, so a period-on-period change is a floor rather than a measurement. Composite visibility scores are unstable enough that we will not quote them alone.
Who this is for, and who it is not
It fits mid-market B2B companies where being recommended matters more than being clicked, where someone can act on findings internally, and where leadership has noticed buyers arriving with opinions nobody in the building supplied.
We will say no when:
- The search foundation is not there. If your pages take a month to index or only render after JavaScript, GEO is the wrong purchase this quarter and we will say so on the first call.
- Nothing can be baselined. Without a recorded prompt set before the work starts, nothing afterwards is provable, and that baseline cannot be recovered later.
- You want a guaranteed position in an AI answer. Nobody can sell that honestly.
- You want software. We are not a platform. If a dashboard is what you need, buy one and keep your money.
A consistent presence on social media is part of how a model judges that an entity is real and worth citing. Not sure which of these is your problem? That is what an SEO audit is for: it gathers the evidence, and the assessment tells you which finding is the cause and which are symptoms of it.
Questions we get asked
Q. What is generative engine optimization, in plain terms?
It is the work of shaping what AI models say about you, including when they name no source at all. Most AI answers cite nothing. The model simply describes your category, your products and your competitors from what it has absorbed. GEO covers the structural work that makes your position legible, the third-party sources models actually trust, and the measurement of what they currently believe.
Q. What is the difference between SEO, AEO and GEO?
Three questions in a stack. SEO asks whether an engine can find you, and is measured in impressions, position and clicks. AEO asks whether an assistant will quote you and credit you, and is measured in citations and brand mention rate. GEO asks what the model says about you when it is citing nobody, and is measured in share of voice and how you are described. They share a foundation, because a page that cannot be crawled or rendered is invisible to all three, and they are measured with completely different instruments. Judging one by another’s numbers is the most common way all three get written off.
Q. Are you a GEO agency or a GEO platform?
An agency, and a deliberately small one. We are not a software company. We use the tracking platforms, including Searchable, where we are a certified partner, and we will tell you which one produced which number. What a platform gives you is a chart showing that visibility changed. What we do is work out which of several possible causes is real, fix it in the right order, and prove the fix reached the crawlers that matter. If you only need the chart, buy the tool and skip us.
Q. Can you guarantee we will appear in AI answers?
No, and nobody can. AI answers are non-deterministic, so the same prompt can produce different sources on consecutive runs. Anyone showing you a guaranteed placement is showing you a screenshot, not a system. What can be done is to remove the structural reasons you are excluded, strengthen the signals models use to decide what to trust, and measure the direction across enough runs for the reading to mean something.
Q. How do you measure GEO, given the answers keep changing?
By sampling rather than by single queries, and by saying which instrument produced which figure. We track a defined prompt set over time, report share of voice, citation counts, the number of distinct pages being used and brand mention rate, and we re-run any claim that something is absent before it reaches you. A single live query is a sample of one. We do not report those as findings.
Q. Does GEO require a large content budget?
Usually not, and the assumption that it does is why many programs waste money. Most of the gap we find is structural: content the model cannot parse, an identity it cannot resolve, or pages it never reached. Publishing more on top of those problems multiplies them. Content matters once the foundation is sound, and by then you usually need less of it than you expected.
Q. Which AI platforms does this cover?
The ones your buyers use: Google AI Overviews and AI Mode, ChatGPT, Perplexity, Gemini and Copilot. The underlying work is largely shared because they are solving the same problem of deciding what to trust. Where they differ is measurement, and no single tool sees all of them well, which is why we name the source of every number.
Q. What if most of what AI says about us comes from sites we do not control?
That is the normal case, and it is the part most GEO offers ignore. Models lean on forums, review platforms, trade publications and marketplaces, and those frequently outrank your own site as a description of you. Part of this work is finding which third-party sources carry weight in your category and deciding, deliberately, which ones are worth influencing and which are not worth the effort.
Q. How is this different from hiring a content agency to optimize our pages?
A content agency writes. This work is mostly diagnosis and structure: whether the engines can reach your pages, whether your identity resolves, whether your schema is served in a way non-Google crawlers can read, and which sources the models trust instead of you. We write when writing is the answer. It usually is not the first answer.
Start with a conversation
Thirty minutes on what AI assistants currently say about you, and whether the gap is structural or reputational. If the honest answer is that you need search foundations first, we will say that rather than sell you the more fashionable service.