The Buyer-Prompt Swipe File: 60 AI Visibility Checks and a Gap Log

Most businesses test AI visibility with five vanity prompts like best [category], then wonder why competitors appear in the answers buyers actually use. I build the prompt set like a PPC search-term list instead: by intent stage, modifier and competitor, with a gap log that matters more than the tool running the checks.

First, know what you are logging

Use this distinction before you copy a single prompt. A brand can appear in an answer without getting a link, and a page can earn a citation without the answer naming its brand. Those are different outcomes and different fixes.

• Mentioned: the answer names your business.
• Cited: the answer links to your page.
• Recommended: the answer tells the buyer to consider or choose you.

The one adjustment that matters: give each outcome its own column. I have seen an answer cite a deep how-to page without naming the brand. That page taught for free and let a competitor close. For a fuller distinction, see this breakdown of mentions, citations and recommendations.

Run the file without fooling yourself

Use this workflow for every block below. Five checks of best [category] tell you who wins one crowded prompt; they miss the problem, pricing and objection questions where a buyer narrows the field. I learned the equivalent lesson in Google Ads: an account can look fine on its head terms while the search terms report bleeds money.

1. Fill the brackets with your real category, buyer and constraints. Use the same wording across engines.
2. Start a clean chat with no history. Run each prompt three times per engine across ChatGPT, Perplexity and Gemini, plus AI Overviews where triggered.
3. Make one gap-log row per run. Record who got cited instead of you, not just whether you appeared.
4. Keep the prompt set stable month to month. Change it deliberately when you learn a new buyer modifier; otherwise, you cannot tell a trend from a different test.

The one adjustment that matters: skip brand-name prompts such as what is [your brand] when measuring discovery. You asked for yourself, so showing up proves little. Category and problem prompts show where you have to compete. I start with Perplexity when I want to inspect citations, but I do not treat one engine as the whole market. Two ChatGPT answers within an hour shared only 54.7% of brands in one check. One run is a mood swing, not a finding.

Block 1: Find the problem prompts that choose sources early

Use when the buyer feels the pain but has no vendor list. These prompts tend to pull explainers and how-tos. If you are absent here, a comparison page alone will not put you into the earlier answer.

1. How do I fix [painful problem] without hiring [expensive role]?
2. Why does [recurring problem] keep happening with [current solution type]?
3. What causes [symptom] in [buyer context]?
4. How do [buyer type] handle [workflow] when [constraint]?
5. What is the simplest way to [job to be done] for [buyer type]?
6. Checklist for diagnosing [problem] before buying [category]?
7. [Problem] troubleshooting steps when [tool] stops working?
8. How to prevent [costly mistake] when [situation]?
9. What should I track to know if [problem] is getting worse?
10. When is it time to replace [manual workaround] with [category]?

The one adjustment that matters: write the problem in buyer words, not your category language. Use missed calls after hours, not lead response solution. If the answer repeatedly cites forum threads and competitors’ help docs but never your page, look at the answer page you are missing before you start counting links. An analysis of 75,000 brands found a 0.664 correlation between branded web mentions and AI Overview visibility, roughly three times the correlation for backlinks. That is a reason to inspect what the web says about you, not a reason to skip writing the page that answers the question.

Block 2: Test the shortlist, not just best [category]

Use when the buyer knows the category and wants candidates. This is the AI-answer version of bidding only on head terms: crowded and worth measuring, but useless as your entire plan.

11. Best [category] for [buyer type]?
12. Best [category] for [constraint]?
13. Best [category] with [must-have feature] in 2026?
14. Top 5 [category] tools for [job], pros and cons?
15. Which [category] is easiest to set up for [buyer type]?
16. Which [category] integrates best with [common tool]?
17. Best affordable [category] alternative for [buyer type]?
18. Best enterprise-grade [category] for [regulated use case]?
19. What is the best [category] for [location] businesses?
20. List [category] vendors that serve [niche vertical]?

The one adjustment that matters: add the modifier your buyers use. Best CRM tells you little. Best CRM for home service firms with 3 offices tells you who appears for a more specific buyer. Run this block after the problem prompts, not instead of them. A changing head-term answer should not dictate your whole content plan.

Block 3: Put the competitor in the prompt

Use when the buyer has a vendor in mind and you are fighting for the next slot. A rival appearing repeatedly on these prompts gives you a sharper content assignment than a generic rank report.

21. [your brand] vs [competitor]: which is better for [buyer type]?
22. [competitor] vs [competitor 2]: pros, cons and pricing?
23. Top 3 alternatives to [market leader] for [constraint]?
24. Is [competitor] worth it for [buyer type], or overkill?
25. What do customers dislike about [competitor]?
26. Migrate from [competitor] to [category]: steps and risks?
27. [category] without [hated feature or limit of leader]?
28. Which is easier to use: [competitor] or [your brand]?
29. [competitor] pricing vs [your brand] for [team size]?
30. If I hate [competitor], what should I use instead?

The one adjustment that matters: for prompts 25 and 30, fill the competitor bracket and leave the complaint in place. The answer has to deal with objections rather than recite a feature grid. What the deck calls a “balanced comparison,” I call the page you should have written six months ago. If you appear only when the buyer types your name, you do not have discovery. You have the AI equivalent of a branded search ad pointing at yourself.

Block 4: Make pricing answers show their units

Use when the buyer has intent and a budget line. These answers can pull pricing pages, reviews and forum complaints. If your pricing is vague, somebody else gets to describe it.

31. How much does [category] cost for [team size or volume]?
32. [your brand] pricing vs [competitor] for [buyer type]?
33. Is [category] worth $[price]/mo for [buyer type]?
34. Hidden fees with [category] vendors to watch for?
35. Cheapest [category] that still does [must-have feature]?
36. [category] cost per [unit]?
37. How do [category] vendors charge: flat, per seat, or % of spend?
38. What affects [category] pricing most in 2026?

The one adjustment that matters: force the unit economics. Prompt 36 is the one I would keep if I could keep only one from this block. Per location, per seat or per booked job pushes the answer toward a number a buyer can use. If Perplexity cites a competitor pricing page three times while yours never appears, inspect the page that should state your number, included scope and who should skip the offer. I used to tell clients to hide pricing to get the demo call. I was wrong. Hidden pricing lets the model quote a review site against you.

Block 5: Ask the question your sales page avoids

Use when the buyer likes the category but does not trust a vendor yet. These prompts surface reviews, cancellation terms, support pages and the complaints a neat feature table leaves out.

39. Is [your brand] legit? Reviews and complaints?
40. [category] horror stories: what goes wrong after signup?
41. How long does [category] take to show results for [buyer type]?
42. Who should NOT buy [category]?
43. [your brand] contract, cancellation and refund terms?
44. Does [category] work if [common objection]?
45. [competitor] support quality vs [your brand]?
46. What happens if I pause [category] after 2 months?

The one adjustment that matters: run prompt 42 word for word. Who should NOT buy [category]? asks for the disqualification a features prompt will not. If forums keep supplying the answer while your own terms or support pages stay absent, state the limitation plainly on your site. Do not skip the test because you dislike what it might show. The log is supposed to find thin proof.

Block 6: Add the filter that decides the purchase

Use when the buyer adds an industry, city, tool stack or team-size constraint. These prompts are the search-term-mining session: each may be narrow, but together they show where your niche offer has room to appear.

47. Best [category] for [vertical]?
48. [category] for [vertical] that integrates with [vertical tool]?
49. Who manages [category] for [vertical] in [city or region]?
50. [category] for [buyer type] with [compliance need]?
51. Best [category] for Spanish-speaking customers in [city]?
52. [category] that works with [CMS]?
53. [category] for [seasonal business] before [peak season]?
54. [category] for multi-location brands with [count] sites?
55. [category] for franchises vs independents: differences?
56. [category] for [buyer type] open evenings or weekends in [city]?
57. Affordable [category] for startups under [headcount] in [city]?
58. [category] that supports [payment, booking or CRM tool]?
59. How do [vertical] firms measure ROI from [category]?
60. [vertical] case study: [category] that cut [metric] in 90 days?

The one adjustment that matters: stack the modifiers. For prompt 49, compare city; city plus vertical; then city plus vertical plus tool. Say you spend $20k/month on home services in Austin. Who manages Google Ads for plumbers in Austin? may surface directories, while adding using ServiceTitan can change which vendors appear. Do not assume the extra word helps you. Log whether it does. That modifier gap is the whole file in miniature.

Read the repeats before you buy another dashboard

Use this checklist when you have enough rows to sort by prompt type and by who got cited instead. I look for repeated outcomes, not the single answer that makes a good screenshot.

• Absent across problem prompts: check whether you have a page that answers the buyer’s question in their language. If not, write it.
• Named but never cited: inspect what third-party pages say about you, then make your own relevant page more specific about pricing units, timelines or cancellation terms where you can.
• Cited in Perplexity but missing in ChatGPT and AI Overviews: check retrieval before rewriting a useful page. Look for crawler blocks and compare what Search Console shows, while remembering that Google pools AI Overview clicks into Web search without a separate filter.

The one adjustment that matters: sort by the repeated rival page, not by the prompt that annoys you most. Manual testing gets you much of the signal without software cost; keep it until logging eats your week. A free spot-check can take about an hour a month, while paid trackers range from $29 a month for 15 prompts to $499 or more. Buy a tool when your time costs more than the subscription. If you want the loop handled rather than another dashboard to interpret, groas tracks prompts, answers and citations, share of voice and the page fixes behind them, with changes logged and a human owning the outcome. No junior manager learning on your budget.

The gap log: the artefact I would keep

Use this after every run. Copy the header into a sheet and make one row per run, not per prompt. A 20-prompt set run three times across ChatGPT, Perplexity and AI Overviews gives you 180 rows. That is enough to look for repeats rather than trust whichever answer you saw first.

PROMPT | ENGINE | RUN # | MENTIONED? Y/N | CITED URL | RECOMMENDED? Y/N | WHO GOT CITED INSTEAD | PAGE THAT SHOULD ANSWER IT | FIX
Best CRM for plumbers with 3 offices? | Perplexity | 1 | N | — | N | Competitor A pricing page, G2 list | /industries/plumbing-crm/ | Rewrite page with per-office pricing

The one adjustment that matters: fill in WHO GOT CITED INSTEAD every time you lose. The 60 prompts are replaceable; any decent operator can rebuild them. The log is what forces honesty when the same competitor pricing page beats you three times while your head-term checks look fine. Sort by that column, inspect the page that should have answered, and fix it first. Stop asking AI if it likes you. Log where it picks someone else.