Prompt Intelligence: the questions buyers ask AI about your category
Prompt intelligence is the discovery, classification and tracking of the questions people ask AI assistants about a category, together with a record of which brands each answer names or cites. In 2026, most buying conversations start with an assistant, and the shortlist it returns is where the decision is made. Asva runs a defined library of your buyers' prompts across ChatGPT, Perplexity, Gemini, Copilot and Google AI Mode on a schedule you set, weekly by default, captures each raw answer, tags every prompt with its intent, and records which brands and sources the engines named or cited in reply. The result is a ranked map of the prompts your brand should be present for, and the gaps where it is missing.
Prompt Explorer
What prompt intelligence is, and why it sits upstream
Prompt intelligence turns scattered buyer questions into a measured, prioritised library you can act on.
A list of prompts on its own is a spreadsheet. What makes it intelligence is the layer on top: an intent tag on each prompt, a record of who the engines recommend, a flag on the prompts where no brand is cited, and a per-engine comparison of the same question. That layer is what every other part of an answer engine optimization programme measures against. Visibility tracking, citation analysis and monitoring all need a defined set of prompts to run, and Prompt Intelligence produces and maintains that set.
The distinction that matters to a buyer is between a branded prompt and a non-branded one. Branded prompts, the ones that name you, show what people already know; the decision happens earlier, on non-branded category questions where no brand is in the picture yet (“which tool suits a twenty-person team”, “is there a cheaper alternative to X”). Prompt Intelligence seeds from the category first, then adds branded prompts as a smaller subset.
How it works, step by step
Prompt Intelligence goes from a category seed to a tracked, scored prompt library in five stages.
- 1
Seed the category
You enter your category, your brand and the competitors you care about, and Asva expands those seeds into the questions buyers ask about the category as a whole, not only the ones that name you. The same category produces different questions in India, the US and the UK, so each region is seeded separately.
- 2
Generate and classify prompts
The product produces a library of distinct phrasings, de-duplicates them into separate prompts, and tags each with an intent, a topic, a keyword and whether it is branded. You review, retag and add your own phrasings from sales calls and support tickets; added prompts track on the same schedule as discovered ones.
- 3
Run every prompt on every engine
Each prompt runs on ChatGPT, Perplexity, Gemini, Copilot and Google AI Mode, in each region you track, on your cadence. The full answer is captured every time, not a screenshot, so mentions, citations, ordering and sentiment are extracted consistently and a change after publishing is measurable against a baseline.
- 4
Score presence and surface the gaps
For every prompt and engine, the product records whether you were named, whether a page of yours was cited, which competitors appeared and in what order, and which sources the answer leaned on. Prompts where no brand is cited, or a rival is cited on a source you could match, become ranked gap prompts.
- 5
Brief the content and track the change
Each gap prompt becomes a content brief stating the question, its intent, the competing sources and the format the engine prefers. After you publish, the same prompts keep running, so presence reads as a rate and you see which engine moved.
Before connecting a brand, the free AEO prompt generator produces a starter set and Prompt Analysis shows citation rates on a sample.
What it measures on each engine
The same prompt is a different question on each engine, because each retrieves from a different index and weights sources its own way.
Running one blended number across all engines hides where the work is. In the 30 days to 18 September 2026, the engines split their citations unevenly across the category Asva tracks: Perplexity 27.95%, ChatGPT 26.64%, Gemini 23.31%, Copilot 11.14% and Google AI Mode 10.94% (source: Asva AEO MCP, 12,849 citations across 1,950 tracked answers). A brand present on three engines and absent on two has a source-coverage problem on those two, not a content problem everywhere, which only per-engine tracking reveals.
ChatGPT
ChatGPT answers from model knowledge unless it decides to search, through OAI-SearchBot and ChatGPT-User. Comparison and purchase-style prompts trigger a search far more often than informational ones, so presence depends on being fetchable at query time. The same prompt can be searched one day and answered from memory the next.
Perplexity
Perplexity retrieves for almost every prompt and cites sources inline, so per-prompt citation data is richest here. It favours pages that answer the specific question directly; keep PerplexityBot unblocked.
Google AI Overviews and AI Mode
Both draw on Google's web index, so results track more closely with classic search. AI Mode handles longer, multi-part prompts, where one intent fans out into many phrasings most visibly. Google-Extended governs training use, not whether you appear.
Gemini
Gemini blends model knowledge with Google Search grounding and names brands readily in informational answers. Comparison prompts often return a structured table, which changes what counts as presence: a row in the table, not a mention in prose.
Microsoft Copilot
Copilot is grounded on Bing, so a site thin in Bing's index is thin here whatever it does elsewhere. It cites with numbered references, and is the engine most often left out of an audit.
The metrics, and how to read them
Every number Asva reports is more useful split by intent and engine than read as a blended total.
Prompt coverage
The share of tracked prompts on which your brand appears at all, on any engine. Read it by intent: informational and transactional coverage are different numbers with different fixes, and the blended figure hides both.
Citation rate
The share of prompts where the answer cites one of your pages, not merely names the brand. A mention without a citation is model knowledge and can vanish when the model updates; a citation is earned by a page and more durable.
Share of voice by prompt
Your mentions divided by all brand mentions across a prompt set, read per engine and per intent. A rising total driven by one engine is not a category-wide gain.
Gap count and gap value
How many prompts return no cited brand, weighted by intent and proxy volume. A falling gap count with flat coverage means competitors are filling gaps faster than you.
Engine spread
The difference between your best and worst engine on the same prompt set. A wide spread almost always points to a crawler or source-coverage issue on the weak engine, to check before rewriting content.
Sentiment
The tone of the answers that mention you, scored per prompt and rolled up per engine, so a rise in mentions with a fall in sentiment is not mistaken for a win.
No metric here is a traffic estimate. No assistant publishes prompt volumes, so Prompt Intelligence uses directional proxies, how many phrasings converge on one intent and how often engines suggest a prompt as a follow-up, to rank prompts against each other, not to imply visits.
How prompts are classified
Each prompt carries an intent tag, because visibility usually differs sharply across intents even for the same brand.
Informational
Wants an explanation (“what is answer engine optimization”). Answered from model knowledge more often, so durable presence means being the definition the model learned.
Comparison
Wants options weighed (“X vs Y for a small team”). Triggers retrieval and cites sources, so a comparison page or neutral listicle can earn the citation.
Transactional
Close to acting, wants a recommendation (“best AI visibility tool for an agency”). The highest-value intent, and usually where brands are weakest.
Problem-solving
Has a specific job to do (“how do I get cited by Perplexity”). Answered by how-to content that states steps a reader could follow.
Best-of / listicle
Wants a ranked shortlist (“top tools for …”). Engines lean on third-party rankings here, so presence often depends on off-site coverage.
Branded, topic, region and constraints
Each prompt also carries whether it is branded or non-branded, its topic, its region, and the constraints it embeds. Every tag is editable, and reporting rolls up by each dimension, so “transactional intent on Perplexity in the US” is a view of its own.
Terms are defined in the AEO glossary, with a dedicated prompt intelligence entry.
Who uses it, and for what
The prompt library is one shared asset; what each team does with it is not.
The SEO or AEO analyst
Turns a keyword-shaped plan into a prompt-shaped one: gap prompts become briefs, engine spread points to the crawler fix, and per-intent citation rate becomes the report that sits next to rankings.
The brand or PR team
Reads the exact questions where an assistant frames the brand and the sources it leans on. Reputation work starts from the prompt and the cited source, and net sentiment per prompt shows where a correction is worth chasing.
The agency
Runs the same library across many client brands and reports each on a common footing, coverage, share of voice and gaps per client, without rebuilding the analysis by hand. Pairs with the agency solution for white-label reporting.
The exec or founder
Wants one honest line: which category questions we win, which we lose, and to whom. The per-intent split turns “how are we doing on AI search” into a number with a fix attached.
Deeper cuts are in the SEO and AEO teams, PR and brand teams, SaaS and e-commerce and retail solution pages.
How it differs from a rank tracker or social listening
Prompt Intelligence measures whether an engine recommends you in an answer, which neither a rank tracker nor a listening tool reports.
A rank tracker records where a page sits on a results page for a keyword, and that position no longer decides the outcome: an assistant composes one answer from the sources it retrieves, and if yours is not among them you are absent regardless of rank. Prompt Intelligence works at the level of the answer: the prompt, the brands named, the sources cited.
Social listening counts mentions across social platforms and news, telling you where you are talked about. It does not run the buyer's question through an assistant, and cannot tell you which brand the engine recommends instead of you, or which source that rests on. Prompt Intelligence starts from the question a buyer asks and ends at the source the assistant trusted, the causal chain a mention count cannot see.
Rank tracker or social listening
- Where a page sits on a results page for a keyword
- Mentions counted across social platforms and news
- Never runs the buyer's question through an assistant
- Outcome: a position, or a mention count
Prompt intelligence
- The prompt a buyer asks, the brands named, the sources cited
- Which brand the engine recommends instead of you, and why
- The same question run on every engine, on a schedule
- Outcome: cited in the answer, per engine
How it connects to the rest of Asva
Prompt Intelligence supplies the prompt set that every other Asva module reports against.
The Brand Visibility Tracker reports presence across that set over time; Citation Intelligence shows which of your pages earned the citations; AI Search Monitoring alerts you when a tracked answer changes; and the AI Traffic Decoder connects the prompts to the referral traffic they send. Where those questions are answered from forum threads, Reddit AI tracking follows the UGC-to-citation path that best-of prompts often rely on.
The same prompt data is available outside the dashboard through the Asva AI MCP server. Its list_topics, list_prompts and get_prompt_responses tools expose your tracked topics, your classified prompt library and the raw AI answers behind any prompt, so you can ask an assistant “list the prompts where we are not mentioned for topic X” in the thread you already work in, backed by your live workspace. Setup and the full tool reference are on the MCP docs page, the Asva AI data MCP that reads your visibility workspace.
Metrics glossary
Short definitions for the terms this page uses.
Gap prompt
A question the engines answer without citing any brand, or citing only a weak source. The highest-leverage target, because there is no incumbent to displace.
Citation rate
The share of prompts whose answer cites one of your pages, not merely names your brand. See citation rate.
Share of voice by prompt
Your mentions divided by all brand mentions across a prompt set, read per engine and intent. See AI share of voice.
Proxy volume
A directional ranking signal for a prompt, used because no engine publishes real prompt volumes.
Intent classification
The tag for what the person wants: informational, comparison, transactional, problem-solving or best-of.
Branded vs non-branded
Whether a prompt names your brand; non-branded category prompts are where the buying decision is made. See AI citation.
Frequently asked questions
Common questions about how Asva discovers, classifies and tracks buyer prompts.
What is prompt intelligence and how does it work?+
Prompt intelligence is the discovery, classification and tracking of the questions people ask AI assistants about a category, with a record of which brands each answer names or cites. Asva seeds a prompt library from your category, runs those prompts on ChatGPT, Perplexity, Gemini, Copilot and Google AI Mode on a schedule, and scores every answer for presence, citations, competitor mentions and sentiment.
How is prompt intelligence different from keyword research?+
Keyword research studies short strings typed into a search box and estimates their volume. Prompt intelligence studies full questions with context, classifies their intent, and measures which brands the assistants name in reply. Keywords tell you what people search; prompts tell you what they ask and who gets recommended.
Which AI engines does Asva track for prompt data?+
Asva runs tracked prompts on ChatGPT, Perplexity, Gemini, Microsoft Copilot and Google AI Mode, across India, the US and the UK, and reports presence, citations and competitor mentions separately for each. Engine and region coverage is selectable per brand.
Does prompt intelligence show real prompt volumes?+
No. No assistant publishes prompt volumes, so any exact count would be invented. The product provides volume proxies, such as how many phrasings converge on an intent and how often engines suggest a prompt as a follow-up, and labels them as directional, to rank prompts rather than imply traffic.
What is a gap prompt?+
A gap prompt is a question buyers ask for which the engines cite no brand, or only a weakly sourced one. They are the highest-leverage targets because there is no incumbent to displace. Asva ranks them by intent and proxy volume and shows what the engine cited instead.
Why does my brand appear on one engine and not another for the same prompt?+
Each engine retrieves from a different index and weights sources differently. The usual causes are a retrieval crawler blocked in robots.txt, content assembled in the browser and invisible to that crawler, or thin coverage in the index the engine is grounded on.
Can I add my own prompts, and query them outside the dashboard?+
Yes to both. The generated library is a starting point; you can add phrasings from sales calls, support tickets and research, tracked and scored the same way, and re-run on your schedule (weekly by default). The same data is reachable through the Asva AI MCP server, so you can list and read prompts and their raw answers from an assistant like Claude or ChatGPT.
See the prompts
your buyers are asking
Bring a category and a competitor list. We will show you the prompt library, the intent split, and the gap prompts nobody is cited for.
Analyze my category