Intent Mapping — How to Define What's Worth Testing in an AI Visibility Audit
Having personas isn't enough. A persona tells you who is asking. It doesn't tell you what they're asking, when in their decision process they're asking it, or what frame they're bringing to the question. All three of those things shape what an AI says in response — including which brands it mentions, which sources it cites, and how it frames the answer.
Intent mapping is the step that defines the full space of questions worth testing before you write any of them. It takes each persona and asks: given what this person is trying to accomplish and what they already know, what are all the meaningfully different conversations they'd have with an AI assistant across their decision journey? The output is a structured map of which scenarios belong in your test set, and which priority they should get.
Without this step, question sets cluster around the middle of the buyer journey — the questions that feel natural to write, usually because they sound like how your marketing team talks about your category. The scenarios that actually reveal the most significant visibility gaps get skipped.
This post covers how to build the map. It continues the previous post about brand discovery and persona design.
The Two Variables That Determine AI Behavior
For any given persona, the AI's response depends primarily on two things: what the buyer is trying to accomplish, and what they already know going in.
These are independent variables. They interact. And changing either one changes the AI response — sometimes dramatically.
What the buyer is trying to accomplish. A buyer figuring out whether they have a problem gets different AI output than a buyer comparing platforms, who gets different output than a buyer who's ready to select a vendor and wants a shortlist. This isn't just a difference in question phrasing. It's a difference in what the AI is being asked to do — diagnose, educate, compare, recommend — and AI assistants handle each of those jobs differently. They cite different sources, reach for different levels of specificity, and name brands at very different rates depending on whether the question calls for that or not.
What they already know — their question context. A buyer describing a problem in plain language, with no category or vendor vocabulary, gets different AI output than a buyer who uses the category name, who gets different output again than a buyer who names a specific competitor. The vocabulary in the question signals the AI about what kind of answer to produce. Strip that vocabulary away — ask from the buyer's raw experience of the problem before they've learned the category — and the AI has to answer from content authority alone, without being pointed toward any particular solution space. That's the hardest scenario for your brand to appear in organically. It's also the most important one to test.
Mapping intent means working through every persona and asking: across their decision journey, what are the distinct combinations of what they're trying to accomplish and what they already know? Each distinct combination is a scenario worth considering for the test set.
Working Through a Persona's Journey
Start with the persona you defined in Step 2 and think through their journey from the earliest moment of AI-assisted research to the latest.
At the earliest stage, many buyers don't yet have category language. They describe their situation — what's broken, what's slowing them down, what's creating pressure — without naming a category or vendor. This is where the most significant visibility gaps often live, because AI models answer these questions from whatever content authority exists in the space, without a vendor frame to anchor the response. Most brands underinvest in content that speaks to this moment.
As buyers move through their journey, they develop vocabulary. They start using category names. They encounter competitor names. They get a shortlist together. At each stage, the AI has progressively more signal to work with — and the response looks progressively more like the kind of recommendation content your marketing team would recognize. These stages are important to test too, but they're not the only stages, and they're often the only ones that informal audits cover.
At the end of the journey, a buyer who already knows your name may ask about you directly — validating whether you fit their criteria, asking what you're known for, checking how you compare to something else they're considering. These scenarios test something different: not whether AI surfaces you, but whether what it says about you is accurate and helpful when it does.
For each persona, trace this arc and identify the distinct moments in it. Not every persona has the same moments. An operations manager responsible for configuring a support platform has early questions about how to audit their current setup — questions that don't sound like platform evaluation at all. A leadership buyer may skip that entirely and start at the comparison stage. What makes the mapping step meaningful is that it's done persona by persona, based on the decision context and primary query surface you defined in Step 2, not from a generic buyer journey template.
Question Context Changes the Response
For each moment in the journey you identify, think through what the buyer's question context looks like — what vocabulary and frame they're bringing to the question.
The most important question context to test for most personas is the one where they describe the problem with no category or vendor language at all. This is harder to write test questions for — it requires genuinely getting into the buyer's head before they know your category exists — but it's the scenario that reveals whether your brand can win a conversation that no one has been pointed toward. A question that uses your category name is already partway to finding you. A question that describes the underlying problem is not.
From there, think about when this persona would start using category language, when a competitor might enter their frame, and when they'd be asking directly about your brand. These are different question contexts for the same underlying intent, and the AI's response shifts with each one. A recommendation-seeking question at an unbranded context produces very different results than the same request phrased with a category name in it — even when the underlying buyer need is identical.
The goal isn't to test every possible question context for every intent. It's to identify, for each scenario you're considering, which context is most plausible for this persona and which context would reveal something the other scenarios don't. For most personas, you want at least one scenario where the buyer has no category vocabulary yet. For commercially important personas, you also want scenarios where a named competitor is in the frame — because those scenarios reveal who's winning the conversations you're losing.
What the Buyer Is Looking For
There's a third variable worth noting, though it's less a standalone dimension and more a modifier to apply per scenario: is this buyer trying to understand something, or trying to get a recommendation?
The same question can go either way. "How do unified support platforms handle omnichannel routing?" is informational — the buyer wants to understand. "What's the best unified support platform for omnichannel routing?" is recommendation-seeking — the buyer wants a name. The AI responds to these differently. Informational questions tend to produce explanation and criteria. Recommendation-seeking questions tend to produce named brands. Both are worth testing, but not for the same personas in the same scenarios — and the distinction changes what a meaningful finding looks like when results come back.
For each scenario in your map, note which mode is most plausible for this persona at this stage. A buyer early in the journey, without category language, is more likely to be asking to understand. A buyer who's named a category and wants a shortlist is clearly recommendation-seeking. In many scenarios, both modes are plausible, and running both reveals different things.
Building the Map
Once you've worked through the journey for each persona — identifying the key moments, the relevant question contexts, and whether the buyer is seeking to understand or seeking a recommendation — you have a candidate set of scenarios. The next step is prioritizing them.
Not every scenario warrants equal coverage. A useful rule: prioritize scenarios where the intent-context combination is both plausible for this persona and likely to produce AI behavior that's meaningfully different from the other scenarios in your set. If two scenarios would produce nearly identical questions and nearly identical AI responses, they don't both need to be in the test.
Beyond that, a few scenarios should almost always be in the high-priority tier regardless of persona: the earliest-stage, unbranded question for any commercially important persona, and the direct brand-validation question for any persona who plausibly has your brand in frame. The first tells you whether you can win a conversation before the category frame is set. The second tells you what AI says about you when it does surface you — and whether that description would help or hurt a buyer who encounters it.
Give each persona enough scenarios to cover the real diversity of their journey, but not so many that each gets only shallow test volume. A practical range is six to eight scenarios per persona. That's enough to see meaningful patterns without spreading test resources so thin that single-conversation outliers drive your findings.
What This Looks Like in Practice
The Freshdesk persona set from Step 2 produced five personas. For each, intent mapping identified the key moments in their decision journey, the relevant question contexts for each moment, and which scenarios warranted high-priority test coverage.
For the Customer Service Leader, the journey starts with unbranded questions about operational fragmentation — before any category language is in frame — and progresses through category comparison, vendor selection, and direct brand validation. The highest-priority scenarios sit at the early and late ends: the unbranded moments where the buyer is describing problems without vendor vocabulary, and the brand-aware moment where they're validating whether Freshdesk fits their criteria. The middle of the journey — category-led comparison and vendor selection — is important but also the most likely place for Freshdesk to appear regardless, because that's where category-association content tends to do its work.
For the Support Operations Manager, the journey looks different. The earliest questions are operational and diagnostic — how do I figure out where my routing is breaking down — and they don't sound like platform evaluation at all. Testing from this persona at that early stage requires questions that are firmly in the problem space, without category vocabulary, and the AI responses will reflect whether Freshdesk has built content authority around support operations fundamentals, not just around its product features.
PERSONA 1: Customer Service Leader Comparing Omnichannel Support Platforms
HIGH PRIORITY
Early-stage, no category frame
The buyer is describing fragmented support operations — disconnected channels,
poor visibility, pressure to scale — without yet reaching for category or vendor
language. No product names, no category terms in the question.
Testing: whether Freshdesk surfaces in responses driven purely by content
authority, before the buyer has a vendor frame.
Early-stage, no category frame — seeking a recommendation
Same unbranded framing, but the buyer asks what kind of tool or approach could
help them consolidate. Still no category or vendor names.
Testing: whether Freshdesk gets named when the buyer asks for help without
pointing toward any specific solution space.
Category-known, understanding tradeoffs
The buyer knows the category name and wants to understand how platforms differ
on scale, analytics, AI, and ease of implementation. Informational, not yet
asking for a shortlist.
Testing: whether Freshdesk appears in evaluation-frame content, and how
accurately it's described.
Category-known, ready to select
The buyer is asking for a recommended shortlist of platforms for unified
channels, ticketing, and AI support. Direct recommendation request.
Testing: whether Freshdesk appears when a buyer with purchase intent asks
for names in the category.
Competitor in frame, asking for alternatives
The buyer knows a major competitor and asks what else should be on their
shortlist or how alternatives compare.
Testing: whether Freshdesk appears when the conversation starts from a
competitor, and which brands dominate that conversation when it doesn't.
Freshdesk in frame, validating fit
The buyer has Freshdesk on their shortlist and is asking directly whether it
fits their criteria: unified channels, AI productivity, legacy replacement.
Testing: accuracy and helpfulness of what AI says about Freshdesk when asked.
---
PERSONA 2: Support Operations Manager Setting Up Routing, SLAs, and Automation
HIGH PRIORITY
Early-stage, operational problem frame
The buyer is describing what's wrong with their support operation — queue
chaos, manual routing, missed SLAs — without using help desk or platform
vocabulary. No category or vendor language.
Testing: whether Freshdesk surfaces in responses to operational problems
before the buyer has a platform-selection frame.
Operational problem frame, seeking practical guidance
The same problem frame, but the buyer wants steps for how to fix it — still
without reaching for product or category language.
Testing: whether Freshdesk appears in how-to responses to implementation
problems, not just in platform recommendation lists.
Category-known, seeking configuration guidance
The buyer uses help desk or customer service software vocabulary and wants
guidance on setting up routing rules, SLAs, and workflow automation.
Testing: whether Freshdesk appears when a buyer with category knowledge
asks implementation-specific questions.
Freshdesk in frame, validating operational fit
The buyer is evaluating whether Freshdesk can specifically handle what they
need: channel-specific SLAs, routing controls, integrations, handoff logic.
Testing: whether AI accurately describes Freshdesk's operational capabilities
in terms that matter to an implementation-focused buyer.
What You Have at the End of This Step
A completed intent mapping step gives you two things the question development step depends on.
A prioritized scenario set. For each persona, a defined collection of scenarios — each specifying the stage of the buyer's journey, the question context, what kind of answer the buyer is looking for, and what to avoid in the question itself. These scenarios are the direct input to question writing in Step 4. Every question you write will be written from one of these frames.
A coverage check. A view of what's in your test set and what's not. The most common gap: heavy coverage of category-known and brand-aware scenarios, thin coverage of early-stage unbranded scenarios and competitor-in-frame scenarios. A test set with that gap will produce results that reflect your visibility after buyers already know your category — not your visibility at the moment they're forming impressions for the first time.
Next in this series: Step 4 — Question Development: how to write test questions in genuine buyer language — and why the instinct to phrase them in a way likely to surface your brand is the fastest route to results that look good and mean nothing.
If you'd rather see what your brand's intent map looks like before working through this yourself, request a diagnostic run.
By Gaurav