LinkedIn
I used to mine the search term report every Monday before I touched a bid. I treat AI answers the same way: run a fixed bank of real buyer questions, log who gets named and cited, then mine the misses like negative keywords.
Swipe 1: Freeze 50 buyer prompts before you check the answers
When to use it: Build the bank before you start monitoring, not after an answer annoys you. Copy the prompts into a document, replace the bracketed fields once, and freeze the wording for 60 days. Sort by intent so you can tell a missed buying question from a flattering branded mention. Backlinko’s prompt-tracking approach makes that distinction; Flow20’s monitoring guide also calls for fixed wording, country and period.
Run category prompts first. These buyers know the service but have not picked a vendor.
CATEGORY PROMPTS — buyer knows the service, not the vendor
1. Best [SERVICE] for [PROBLEM] in [CITY] for a [BUDGET RANGE] budget
2. Best [SERVICE] for [AUDIENCE] who need [OUTCOME] fast
3. Most reliable [SERVICE] near [CITY] with good reviews for [PROBLEM]
4. Top-rated [SERVICE] for [PROBLEM] that offers [KEY FEATURE]
5. Affordable [SERVICE] in [CITY] that handles [SPECIFIC JOB] without upsells
6. Which [SERVICE TYPE] is right for [PROBLEM] in a [HOME TYPE / BUSINESS TYPE]?
7. Best [SERVICE] for [PROBLEM] when [CONSTRAINT]?
8. Who does [SPECIFIC JOB] in [CITY] with upfront pricing?
9. Best-reviewed [SERVICE] for [PROBLEM] according to recent customer reviews
10. What should I look for when hiring a [SERVICE] for [PROBLEM]?
The one adjustment: Use the geography you actually serve. A generic best [SERVICE] prompt tells you less about the shortlist a local buyer sees than best [SERVICE] for [PROBLEM] in [CITY].
When to use problem prompts: Your branded searches look fine, but new buyers are not finding you. I used to tell clients branded search proved a brand was healthy. I was wrong. People can choose someone else before they ever learn your name.
PROBLEM PROMPTS — buyer has pain, no vendor in mind
11. How do I fix [PROBLEM] without [COMMON FEAR]?
12. Why does [PROBLEM] keep coming back after [COMMON FIX]?
13. How much does it cost to fix [PROBLEM] in [CITY] in 2026?
14. How long does [SERVICE JOB] take for a [HOME TYPE]?
15. Is [SYMPTOM] an emergency or can it wait until morning?
16. What causes [PROBLEM] in [SEASON] and how do I stop it?
17. DIY vs hiring a pro for [PROBLEM]: what actually breaks when I DIY?
18. How do I choose between repair and replace for [ITEM]?
19. What questions should I ask before hiring someone for [PROBLEM]?
20. How do I stop [PROBLEM] from voiding my [WARRANTY / INSURANCE]?
The one adjustment: Keep the buyer’s fear in the question. Without tearing up the yard is not interchangeable with fast. One describes the objection you need to answer.
When to use comparison prompts: You keep hearing we went with someone cheaper, but you never got to pitch. These questions expose the tradeoffs on the shortlist.
COMPARISON PROMPTS — buyer is shortlisting
21. [YOUR SERVICE] vs [ALTERNATIVE] for [PROBLEM]: pros and cons
22. [COMPETITOR A] vs [COMPETITOR B] for [SERVICE] in [CITY]: which is cheaper?
23. [YOUR CATEGORY] vs [ADJACENT CATEGORY] for [PROBLEM]: which lasts longer?
24. Is [PREMIUM OPTION] worth it over [BUDGET OPTION] for [PROBLEM]?
25. Cheapest way to fix [PROBLEM] that still passes [INSPECTION / WARRANTY]?
26. [SERVICE] pricing in [CITY]: flat rate vs hourly, what is fair in 2026?
27. Best alternative to [WELL-KNOWN COMPETITOR] for [PROBLEM]?
28. [BRAND CATEGORY] with [FEATURE] vs without: does it matter?
29. Local [SERVICE] vs national chain for [PROBLEM]: who shows up faster?
30. What is the catch with cheap [SERVICE] quotes for [PROBLEM]?
The one adjustment: Put the rival you actually see in prompt 27. A generic competitor placeholder produces a generic alternatives list.
When to use brand checks: After the unbranded money prompts. These tell you whether an engine knows you exist and gets your details right. They do not tell you whether it would recommend you to someone who has never heard of you.
BRAND-CHECK PROMPTS — does the engine get your business right?
31. What does [YOUR BUSINESS NAME] do for [PROBLEM] in [CITY]?
32. Where is [YOUR BUSINESS NAME] located and what areas does it serve?
33. How much does [YOUR BUSINESS NAME] charge for [CORE SERVICE]?
34. What do reviews say about [YOUR BUSINESS NAME] for [PROBLEM]?
35. Does [YOUR BUSINESS NAME] offer [KEY FEATURE]?
36. Why did you recommend [COMPETITOR] over [YOUR BUSINESS NAME] for [PROBLEM]?
37. Which is better for [PROBLEM] in [CITY]: [YOUR BUSINESS NAME] or [COMPETITOR]?
38. List three alternatives to [YOUR BUSINESS NAME] for [PROBLEM] near [CITY]
39. Is [YOUR BUSINESS NAME] open [WEEKENDS / EVENINGS] for [PROBLEM]?
40. What is the phone number and booking link for [YOUR BUSINESS NAME]?
The one adjustment: Run prompt 36 as a follow-up when the engine recommends a competitor, in that same chat. Ask why. The explanation may show you which source or detail influenced the answer; this small-business audit guide uses the same follow-up.
When to use local urgency prompts: You serve buyers who need a job done soon, not just people researching options. Keep these separate from brand checks even though they finish the bank: they do not name you.
LOCAL URGENCY PROMPTS — buyer needs a workable next step
41. Best [SERVICE] open now near [ZIP] for [URGENT PROBLEM]?
42. Who fixes [PROBLEM] with a warranty in [CITY] under [$X]?
43. [SERVICE] for [LANDLORDS / RENTERS / SMALL BUSINESS] in [CITY]: who handles paperwork?
44. How do I book [SERVICE] for [PROBLEM] this week in [CITY]?
45. What happens during a [SERVICE VISIT] for [PROBLEM]?
46. How do I know if a [SERVICE] quote for [PROBLEM] is fair?
47. What [LICENSE / INSURANCE / GUARANTEE] should a [SERVICE] have in [STATE]?
48. How fast should [SERVICE] respond to [URGENT PROBLEM]?
49. Best time of year to book [SERVICE] for [PROBLEM] to save money?
50. What maintenance stops [PROBLEM] coming back after [SERVICE]?
The one adjustment: Replace every timing, price and availability placeholder with a condition a buyer could actually ask about. Otherwise you are testing your template, not your market.
Swipe 2: Log the answer, not your memory of it
When to use it: Start the sheet before the first run. Answers can change between sessions, so a screenshot of one favorable list is not a trend. Backlinko logs prompts, mentions, citations, competitors and dates; Flow20 separates mentions, citations and accuracy. I would rather have an untidy row than a polished claim that “ChatGPT knows us.”
Paste as row 1 in Sheets. Use one row per prompt x engine x check date.
Prompt # | Prompt text (frozen) | Category | Engine | Date | Location | Model/version | Run 1 mention? | Run 2 mention? | Mention rate % | Cited? (Y/N + URL) | Position in answer (1st / mid / last / list #) | Competitor named | Sentiment (+ / = / -) | Facts wrong? | Source that won the citation | Fix needed (page / block / outreach)
The one adjustment: Keep mentioned and cited in separate columns, and record the URL rather than just a yes. A mention tells you that your name appeared. A citation tells you which page the answer surfaced. Neither is a click or a sale. If you appear in one of two runs, log 50%, not “we made the list.”
Swipe 3: Run a 30-minute weekly triage
When to use it: You need a repeatable check without turning prompt monitoring into another account to manage. For this short routine, run 10 prompts on one engine and repeat each twice. Rotate through the bank and the engines; do not call this a weekly audit of all 50 prompts across all three engines. A complete sweep takes longer.
0:00–0:05 Reset. Open clean chats, turn off memory, and note engine, model/version, location and date.
0:05–0:20 Run 10 frozen prompts on one engine, twice each. Rotate prompt numbers and engine on the next check. Log the answers.
0:20–0:25 Score each prompt: mention rate, cited URL, position, competitor named, facts wrong.
0:25–0:30 Pick ONE miss: a problem or comparison prompt with no mentions that maps to a page you can edit.
Rule: never rewrite a prompt to get a nicer answer. Fix the page, not the question.
The one adjustment: If a shift looks important, repeat that prompt across clean sessions before you trust it. Best and comparison answers can move between runs; model routing, memory and retrieval can also change what appears. This guide to answer variance explains why, and a controlled tracking protocol calls for 5–10 repeats when testing a change. Two runs are triage, not proof.
Swipe 4: Give a missed prompt a page-ready answer
When to use the definition block: A problem prompt misses you, and the answer cites another service page. Put a direct answer near the top of your relevant page, with the scope and conditions a buyer needs to judge it. The Perplexity page-structure guide recommends direct early answers and self-contained Q&A blocks. Make the useful passage easy to find without pretending a block guarantees a citation.
DEFINITION BLOCK — adapt for the service page behind prompts 11–20
[Service] for [Problem] in [City] costs [$X–$Y] in 2026 and takes [N hours].
"[Your Business] charges [$X] for [core job], including [scope item 1, scope item 2, warranty]."
Best for: [audience + situation]. Skip if: [situation where you are wrong].
Book: [phone] / [link]. Serves [areas]. Updated [Month Year].
The one adjustment: Replace the price, time, scope and date with details you can stand behind. If you cannot support a number, do not publish the placeholder as a claim. The answer needs to help a buyer, not merely look quotable.
When to use the comparison block: A shortlisting prompt cites a competitor, and your page never states who should choose your option. Give the reader the tradeoff in one place rather than making them reconstruct it from sales copy.
COMPARISON TABLE BLOCK — adapt for the page behind prompts 21–30
| Option | Cost in [City] | Time | Lasts | Pick it if |
| --- | --- | --- | --- | --- |
| [Your service] | [$X] | [N hrs] | [N yrs] | [situation] |
| [Alternative] | [$Y] | [N hrs] | [N yrs] | [situation] |
WHO IT'S FOR / WHO SHOULD SKIP IT — place under the table
Choose [Your Business] if: [reason 1, reason 2, reason 3].
Skip us and call [alternative] if: [reason 1, reason 2].
The one adjustment: Make the “skip us” line real. If the cheaper option fits one situation, say which one. That makes your recommendation more useful for the buyer who actually needs you; it does not require handing the whole comparison to a rival.
Swipe 5: Ask for one correction to a cited source
When to use it: The sheet shows the same third-party page winning a citation for three weeks, and your own page has a newer, narrower or better-scoped detail. Send one note. No sequence, no “just bumping this to the top of your inbox,” no campaign dashboard for a single email.
Subject: quick update for your [TOPIC] page
Hi [Name], saw you cite [OLD URL] for [CLAIM] in [YOUR ARTICLE TITLE].
Current for [CITY] in 2026: [YOUR NUMBER, SCOPE AND TIMING].
You can quote this if useful: "[25-word line with the number and scope]."
Source: [YOUR SERVICE PAGE URL, with date].
No worries if you keep what you have.
[Your name, Business, Phone]
The one adjustment: Send it only if the correction is verifiable on the page you link. A dated price with a clear scope gives someone something to check. “We are the better choice” gives them an email to delete.
Swipe 6: Decide when the hand-run sheet is enough
When to use it: Your triage keeps producing more misses than you have time to investigate. The 30-minute routine above is a sample; running all 50 prompts twice across ChatGPT, Perplexity and AI Overviews is 300 answers per complete sweep. Reading and acting on those answers is the expensive part.
KEEP IT MANUAL IF:
- You can run 10 prompts a week and edit one page from what you find.
- You need to learn which buyer questions belong in the bank.
- The sheet still gives you a clear next action.
RETHINK THE WORKFLOW IF:
- You need the full bank checked across all three engines more often.
- The misses outnumber the pages you can fix.
- Monitoring has become the job instead of helping you do the job.
The one adjustment: Price the work of fixing misses, not just collecting answers. You can run your time against whether paying for prompt tracking pays off. I am biased here: groas runs the tracking continuously and then ships the fix. A tracker alone can tell you a prompt missed; somebody still has to rewrite the page. If you need more formats for that work, take the five additional copy-ready page blocks for AI answers.
If I could keep one artefact from this file, I would keep the tracking sheet. The prompts give you questions and the blocks give you a way to answer them, but the row forces the decision: mention rate, cited URL, competitor and next fix in one place. That is what tells me what to write next Tuesday.