Persona gaps in AI answers: why you are visible to some buyers and not others

The dashboard says you're named in a healthy share of answers. The sales team says prospects keep telling them an AI recommended someone else. Both can be true. The dashboard is averaging across a generic asker, and the person who signs the contract isn't generic.

AI answers change with who seems to be asking. A prompt phrased as a finance director gets a different shortlist than the same question phrased as a developer or an office manager. If your visibility is strong for the average prompt and weak for the buyer who matters, the aggregate number hides the problem.

What a persona gap looks like

Take a made-up example for illustration, so no one mistakes it for a result. A project-management tool might show up when someone asks "best project management software", and fall away when the prompt is "project management for a 12-person construction firm that needs to track subcontractors". The generic prompt favors well-known names. The specific one favors whoever has content for that situation.

Gaps typically appear in one of three forms:

Missing entirely. You're absent for a persona that your product serves.
Present but ranked low. Named, but after several competitors.
Present with the wrong framing. Named as a tool for a different buyer, such as "good for enterprises" when you sell to small teams.

The third form is the one people miss, because a mention counts as a success on a dashboard.

Step 1: write the personas you'd actually bother with

Don't start with eight. Pick two or three buyer types who either sign, influence or use the product, and write each as a short description: role, company size, the problem they're trying to solve, what they care about.

Then write six to ten prompts per persona the way that person would phrase them. A finance lead asks about cost, controls and reporting. A developer asks about API, integrations and docs. If your prompts all read like the same marketer wrote them, the exercise will tell you nothing.

Keep a baseline too: the same topics, phrased for no one in particular. That's your control.

Step 2: compare against the baseline

Run each set in a few engines with search on, a few times across several days. For each persona, record:

how often you're named,
where you appear in the list,
how the answer describes you.

Then lay each persona next to the baseline. The biggest difference is where to look. It's more informative than any absolute number, since you're comparing like with like.

A caution: run counts matter. Answers vary from run to run, so two or three runs per prompt are the minimum before you believe a gap. If you only ran each prompt once, treat the result as a lead, not a finding.

Step 3: find the topic behind the gap

A gap for a persona usually traces to one or two topics. The finance lead doesn't see you because you have no pricing transparency, no security page and nothing on reporting. The developer doesn't see you because your docs aren't crawlable.

Look at which topics are weakest for the lagging persona, and check the sources the answers cite. Are those sources pages you could appear on, or content you could write? Our post on personas in AI prompt research explains how to build prompts that reflect the asker, and the persona-based tracking roundup compares tools for it.

Step 4: fix the content, then re-measure

Content for a specific persona answers that persona's real questions. A page titled "for finance teams" that repeats the homepage doesn't. Write the page that a finance lead would send to a colleague: the controls, the audit trail, what it costs and what's included.

Then run the same prompt set again after a few weeks. If the gap narrows, you've learned something. If it doesn't, the cause is probably sources, not your page, which sends you back to step 3.

Where Promptwatch fits

The manual process is sound and slow. Running ten prompts for each of three personas in several engines, several times, then comparing them in a sheet, takes a full day each round. That is why we'd move it into a platform, and the one we'd pick is Promptwatch.

Promptwatch has tracked personas for a while, including persona stacking, which combines persona attributes in a single prompt run. What's new, according to the Promptwatch changelog, is a Persona insights page, added October 23, that compares how your visibility, position and sentiment change per persona.

It's built around the exact comparison from step 2:

Each persona gets a radar chart against the general audience, defined as the answers that ran without a persona. You can see which audience AI recommends you to least.
You can use topics or AI models as the spokes of the radar, so a gap shows as a dent in a specific topic or a specific engine.
A card points out the persona with the biggest gap and its weakest topic or model, which is step 3 handed to you.
Persona is now a heatmap axis, and the persona radar is available as a custom dashboard widget and in PDF reports.

That last point helps with the internal argument. A radar chart in a PDF is easier to put in front of a sales lead than a spreadsheet of run counts.

The changelog doesn't state a plan for Persona insights, so we won't name one. Check the pricing page for what's included.

What it can't tell you

A radar shows where visibility differs. It can't tell you why, and it can't write the page. The diagnosis still depends on your reading of the topic and the cited sources. And the personas are your own construction: if you define them badly, the chart will faithfully measure the wrong people.

Be careful with small samples. A persona with only a few prompts can swing with a single answer. Keep prompt sets large enough that one odd run doesn't change the picture, and don't present a radar to leadership before you've checked that.

For the setup side, see how to track AI visibility for different buyer personas. The short version of this post is: if your numbers look good and your pipeline doesn't, split the numbers by buyer before you change anything else.