RECOMMENDATION QUERY
It does not consult a ranking table. Understanding how a recommendation-shaped answer gets assembled changes what there is to fix.
SHORT ANSWER
Ask an AI to "recommend an answer engine optimization agency" and it does not consult a ranking table. There is no first or second place in an AI answer. A recommendation-shaped query builds its candidate list from third-party documents that line several companies up against the same criteria, then fills in the descriptions from each company's own pages. So appearing on the list needs both halves at once: inside the site, so a machine can read what kind of company this is, and outside it, documents that cover us against the same criteria.
PEOPLE ASK IT THESE WAYS
QUERY TYPE
The single test in this table is whether one document of ours can produce the answer by itself.
| Query type | Example | What it needs | One document enough? |
|---|---|---|---|
| Explanatory | "What is AEO?" | A document where the definition completes within one paragraph | Yes One good explainer of ours can be cited as it stands. |
| Comparative | "What is the difference between AEO and GEO?" | A document with the comparison axes laid out | Yes One document that sets the axes first is enough. |
| Recommendation | "Recommend an AEO agency" | A third-party document listing several candidates against the same criteria, plus each candidate's own description | No However well we describe ourselves, one document of ours does not produce a list. |
HOW IT IS BUILT
This is the structure of the answer, not a measured statistic — anyone can confirm it by asking.
It builds a candidate list
Outside the siteNames are pulled from documents that treat several companies against the same criteria. A page where one company introduces itself cannot produce a list, so most of the material for this step lives outside your site.
It describes each candidate
Inside the siteIt reads the selected companies' own pages and introductions to fill in what kind of place each is. This is where our documents get used — but only after making the list.
It splits by condition
Inside the siteIt closes with situational recommendations — "A if you are an enterprise, B for the domestic market". So if the documents do not say which situation we suit, we get filtered out at the end even when we are on the list.
FOUR REASONS
Three are inside the site and one is outside. Erase that division and this becomes just another checklist.
If the title and first sentence do not say "what kind of company is this", AI has no basis for deciding which list to put us on. However long the service description is, it is not what gets used for classification.
Citable content design →The company description is on screen but missing from the structured data, or the two say different things. AI looks at the machine-readable facts before it reads the screen.
Structured data →AI does not cite a document whole; it lifts one paragraph. Writing that only makes sense after reading the paragraph before it has no fragment to lift, and slides out of the candidate pool.
AEO audit →The candidate list for a recommendation-shaped query usually comes from outside your site — industry lists, comparison articles, press, published data. Absent from those, there is no entrance to the list however much the site is improved.
GEO — becoming the material →OUTSIDE
We do not go in the direction of buying links. Only building material there is a reason to cite.
Places that already line several companies up against one standard. It is the first shape a recommendation-shaped answer reaches for when picking candidates, and being listed or not becomes being on the shortlist or not.
What we do here — Finding the places we can register and filling the information in accurately. That is a different thing from buying a position.
Pieces where a third party compares several agencies. Not a kind we can write ourselves — and were we to write one, being our own article it would not serve as the basis for a candidate list.
What we do here — Publishing our measurement conditions and raw data so that whoever does the comparing can cite them.
Records where journalists or trade press covered a company. They last, and they often become the starting point other documents cite in turn.
What we do here — Building the primary material a journalist could work from — not manufacturing a track record we do not have.
Material we measured ourselves and published down to the conditions. It only actually gets cited if the citing side can verify it.
What we do here — Something we already do. It is why every measurement of ourselves is published.
WHAT WE MEASURE
There is no position to record. These four are what there is.
The denominator and the session conditions are set out in the measurement methodology.
FAQ
There is work that raises the odds of appearing, but appearing cannot be guaranteed. Generative AI produces different answers to the same question depending on the model, the moment and the conversation context, so no fixed slot exists. Instead of a guarantee, we work by recording mention rate and citation rate from repeated measurement under the same conditions, so change can be confirmed.
Top placement is a phrase that only holds when there is an ordered list. An AI answer generates one piece of text at a time; there is no first or second position. That said, people genuinely search using that phrase, so we do not dismiss it — we translate what it is actually aiming at into appearance and citation inside the answer.
For recommendation-shaped queries, usually not. "Recommend some" is a request to list several companies, so AI has to secure a candidate list first, and that list often comes not from a document where one company describes itself but from third-party documents treating several companies against the same criteria. Your own pages fill in the description after making the list. So inside and outside the site have to be handled together.
Whether they publish their measurement conditions, rather than whether they promise visibility. Which answer engines, how many times asked, what the denominator is, and whether the raw data is published. Those are the criteria.
They overlap but the goals differ. SEO puts a document into an index, AEO makes the document a candidate for the answer, and GEO makes the brand material for the answer. A recommendation-shaped query is the last stage, the one that needs all three layers.
There is no way other than asking the answer engines directly and recording it. Whether you are indexed but not cited, or cited but not making the recommendation list, changes the place to fix completely. Our free audit records these stages separately and sets out the order to fix them.
Send us a URL and the free audit asks real questions in all 7 answer engines, recording the four stages separately so the place to fix is marked.
We reply within one business day.