Answer engine optimisation has a data problem. Google publishes query volumes through its ads ecosystem; ChatGPT, Claude, Gemini and Perplexity publish nothing. So when people look for keyword research tools for AEO in LLMs, they are looking for something that does not fully exist yet — and the honest response is to explain what can be measured today rather than sell a keyword volume number nobody actually has.
Here is the current state: what is knowable, what is inferred, what is marketing, and how to do useful research anyway.
Why LLM keyword volume cannot be measured the way Google’s can
Three structural reasons. First, no provider exposes a query index. There is no equivalent of Keyword Planner for ChatGPT, and nothing suggests one is coming.
Second, prompts are not keywords. A Google query averages a few words; a prompt to an assistant is often a paragraph, includes context, and continues across turns. The unit of demand is a conversation, not a string, which breaks the entire volume-per-phrase model.
Third, the answer is generated rather than retrieved. Two people asking the same question get differently worded answers, potentially citing different sources, and the same question asked twice can vary. There is no stable ranked list to measure position within.
Any tool claiming exact LLM search volume for a phrase is modeling or extrapolating, usually from Google data. That can still be useful — but know what you are buying.
What can genuinely be measured today
Four things, and they are more useful than they first appear.
- Brand mention presence. Ask an assistant a set of relevant questions and record whether you are named. Repeat on a schedule and you have a real trend line.
- Citation presence. In systems that show sources — AI Overviews, Perplexity, some ChatGPT modes — you can see whether your URLs are being cited.
- AI Overview presence per query. Whether a Google query triggers an AI Overview at all is observable, and it materially changes how much organic click share remains.
- Referral traffic. Visits from assistant domains show up in analytics. Volumes are small for most sites, but the direction of travel is worth watching.
SEO Rocket sits in the first of these: it counts brand mentions across ChatGPT, Google AI Overviews, Gemini and Perplexity and shows the real example questions that produced them, with no setup required. That gives you a baseline and a trend. Competitor share-of-voice — what proportion of answers a rival owns versus you — is on the roadmap and not shipped, so we will not pretend otherwise.
How to research answer-engine demand without volume data
Stop looking for a volume column and start building a question inventory. The unit of AEO research is the question your customer asks an assistant, and you can collect those from sources that already exist.
Mine People Also Ask boxes for every priority topic — they are Google’s own model of follow-up questions and map closely to how people prompt. Read your support tickets and sales call notes, which contain the exact phrasing customers use when they do not know the jargon. Pull question-form keywords from a conventional index, since “how,” “why,” “which” and “is X better than Y” queries still appear in Google data. And check community sources where people ask in full sentences.
Two hundred questions collected this way beats any speculative LLM volume metric, because they are real questions from real customers.
Testing: the closest thing to a keyword tool right now
The practical technique is manual and works. Take your top 30 questions, ask each one in ChatGPT, Gemini, Perplexity and Google, and record four fields: were you mentioned, who was mentioned instead, what sources were cited, and how the answer framed the topic.
Do this monthly and you have a genuine AEO dataset for your niche. It takes an hour or two, and it tells you specifically which competitors the models consider authoritative and what content those citations come from. That is far more actionable than a modeled volume figure. Automate the mention-counting side with a tool if you can, but do the qualitative read yourself — the framing of the answer is often where the insight sits.
What appears to influence whether models cite you
Nobody outside the labs knows the mechanism, and the research that exists is early and contested. What follows is inference from observation, not established fact.
Retrieval-based systems — AI Overviews and Perplexity most clearly — pull from live search results, so conventional ranking still matters a great deal. Being on page one is close to a prerequisite for being cited. Models appear to favour content that answers a question directly and early, in clear declarative sentences that survive being lifted out of context. Consistency of facts about your brand across the web seems to help, since contradictory information gives a model no confident answer to reproduce. And being referenced on third-party sites — comparison pages, roundups, forums — appears to matter more than for classic SEO, because models synthesise across sources rather than ranking one.
What is not established: whether schema markup helps citation, whether llms.txt files influence anything, and whether there is a distinct ranking system at all versus reuse of existing search infrastructure. Treat all three as open questions.
Choosing tools without overpaying for a young category
The AEO tooling market is roughly a year old and priced optimistically. Some dedicated platforms charge more than a full SEO suite to track mentions across a handful of assistants.
A reasonable stance: do not buy a standalone AEO tool as your first purchase. Get conventional keyword research, competitor analysis and rank tracking working first, since retrieval-based answers lean heavily on organic visibility. Add mention tracking as a component rather than a separate subscription — SEO Rocket includes it alongside keyword research, site explorer, content gap, audits and rank tracking at a flat $50 a month, which is a sensible way to hold a baseline without betting a budget on an unproven category. If you have a large brand-visibility budget and need deep competitive AEO analysis today, a specialist tool may genuinely serve you better; that is an honest recommendation, not a hedge.
A twelve-week AEO research programme
Weeks one and two: build the question inventory from PAA boxes, support tickets and question keywords. Weeks three and four: run the manual test across all four surfaces and record a baseline for your 30 priority questions. Weeks five to ten: publish or rewrite content that answers the questions where you are absent, leading with a direct answer in the first hundred words and using literal question headings. Weeks eleven and twelve: re-test the same 30 questions and compare.
Expect modest movement. Assistants update their retrieval and training on their own schedules, and a page published in March may not surface in answers until well into the summer. Meanwhile, everything in that programme also improves conventional rankings, which is the reason to do it even if the AEO thesis turns out weaker than expected.
The category will mature. Real keyword research tools for AEO in LLMs will eventually offer something closer to demand data, and the first honest ones will be explicit about their methodology. Until then, question inventories, manual testing and mention tracking are the working toolkit — and the sites that keep publishing genuinely useful, well-structured answers are positioned for whatever the measurement layer turns into.
