Skip to content
Chalk Labs / AI SEO
How-to guide

How to Build a Prompt List for AI Visibility Tracking

By Rahil Jain, founder of Chalk LabsUpdated How we research

The short answer

Build a prompt list of about 30 questions, split into five buckets: customer problems, category discovery, comparisons, product selection and brand questions. Write each one the way a real buyer would, add two or three buyer types, and report branded prompts separately because they inflate your score. Keep the core set fixed for at least a month so trends mean something.

  • Start with about 30 prompts: six in each of five intent buckets, written in buyers' own words.
  • No major AI platform except Microsoft Copilot shares prompt volumes, so choose prompts from customer language, not from a volume number.
  • Report branded and non-branded prompts separately, because branded prompts usually inflate the overall rate.
  • Freeze the core set for a month or more. Change prompts only at the bottom of the list, with a start date.
  • Daily tracking adds little for most teams. A fixed list run on a steady schedule matters more.

What makes a good prompt list for AI visibility tracking?

A good prompt list is a short, fixed set of questions that mirror how your buyers ask AI assistants for help, run again and again so you can see change. It is a sample, not a census. Its job is to show whether you are named, cited and described correctly as your work takes effect.

That makes it different from a keyword list. A keyword list tells you what to write. A prompt list tells you whether the AI answers you wrote for are working. If you also need a place to log the results, the tracking spreadsheet guide covers the tabs and formulas. This page covers the step before it: deciding what goes in the first column.

Do it step by step

0 of 7 done0%

  1. Gather 20 to 30 exact phrases from sales calls, support tickets, enquiry forms, reviews and forum threads. Copy them word for word into a sheet. Do not tidy them yet.

    Tip: Search your inbox for the words “do you”, “how much” and “which” to find real questions fast.

  2. Write each buyer type in one line: role, size of business, main worry. These become the voices for your prompts.

  3. Fill the five buckets: problems, category discovery, comparisons, product selection, brand questions. Write each as a full sentence with one detail such as size, budget or place.

  4. Read the list and delete any two prompts that would produce the same answer. Replace them with a different need or a different buyer type.

    Tip: If two prompts differ by one word, keep the one closer to real customer wording.

  5. Add a column that marks each prompt as branded or non-branded. You will report the two groups on separate lines.

  6. Paste five prompts into the assistants you care about. Check each reads naturally, gets a useful answer and does not name you only because your brand is in the question.

  7. Lock the core 30. Pick a run day (monthly by hand, or the schedule your tool offers). Add any new prompt at the bottom with the month it started.

Your ticks are saved in this browser only.

Copy-ready templates

Prompt list sheet (CSV header and sample rows)

text
id,bucket,buyer_type,branded,prompt,start_month
P01,problem,owner,no,"I'm six months behind on VAT for my restaurant, what should I do first?",2026-11
P02,category,owner,no,"Who are the best bookkeepers for restaurants in Leeds?",2026-11
P03,comparison,manager,no,"Local bookkeeper vs online app for a restaurant with 15 staff",2026-11
P04,selection,manager,no,"What should I ask a bookkeeper before hiring one for a restaurant?",2026-11
P05,brand,owner,yes,"Is [Firm name] a good bookkeeper for restaurants, and what do they charge?",2026-11

Paste into a new sheet. The branded column lets you filter and report non-branded prompts on their own.

Prompt-writing formulas by bucket

markdown
- Problem: "I have [problem] at my [business type]. What should I do?"
- Category: "Who are the best [providers] for [buyer type] in [place]?"
- Comparison: "[Option A] vs [Option B] for [size] [business type]"
- Selection: "What should I check before choosing a [provider] for [need]?"
- Brand: "Is [Brand] good for [buyer type], and what does it cost?"

Fill the brackets with detail from real customer wording. Change one bracket at a time to make a new prompt, not five.

Why can't you pick prompts by search volume?

Because nobody outside the AI companies can see prompt volumes. Ahrefs states that, apart from Microsoft Copilot, no major AI platform shares its query data, and that its own AI volume figures are a proxy built from Google search volume, to be used directionally 5. Any tool that shows an exact monthly volume for a long chat prompt is estimating.

Search also happens in fewer chats than people assume. Semrush's clickstream study found web search was switched on for 34.5% of ChatGPT queries in February 2026 8. Ahrefs found only 14.4% of 66.6 million responses it tracked in free, logged-out ChatGPT sessions in August and September 2026 contained fan-out queries, and those averaged about 1.6 fan-out queries each 3. A fan-out query is one of the extra searches an AI runs behind the scenes to build its answer. Google says its AI Overviews and AI Mode may do the same across subtopics 6.

The practical rule: pick prompts where a buyer is likely to be choosing, comparing or fixing something. Those are the questions where an AI is more likely to look things up and name sources. Our query fan-out research guide shows how to see which searches sit behind a prompt.

Which five intent buckets should the list cover?

Use five buckets that follow the buyer's journey: customer problems, category discovery, comparisons, product selection and branded questions. Semrush recommends this spread rather than clustering prompts on brand terms. Its own walkthrough used 30 prompts, six in each of five groups, and advises growing each group to 15 or more only if you need dependable group-level reporting 1.

G2's March 2026 survey of B2B decision makers shows why the middle buckets matter. Their first AI prompts when researching software were 33% category-based, 31% competitor-based, 22% about requirements or process, 9% about the ecosystem and 6% about budget 2. That is software buying, so check whether your market looks the same before copying it. Our own suggested starting split below is equal, because we hold no data on your buyers.

Five buckets, one example each, and what each one tests
BucketExample promptWhat it tells youStart with
Customer problemsWhy is my invoice approval so slow?Whether you appear when the buyer has not named a solution yet6
Category discoveryWhat is the best invoicing software for a small agency?Whether you are on the shortlist6
ComparisonsTool A vs Tool B for a 10-person teamWhether you win when buyers weigh named options6
Product selectionWhat should I check before buying invoicing software?Whether your criteria shape the buyer's checklist6
Branded questionsIs YourBrand any good for agencies?Whether the AI describes you correctly6

How do buyer types and wording change the list?

Each prompt should carry the detail a real buyer adds: their size, budget, location or constraint. A prompt like “best CRM” is vague. “Best CRM for a 5-person recruitment agency in Manchester under £50 a user” is how people actually type, and detail like this gives the assistant more to work with, so test it against your own results.

Pick two or three buyer types and rewrite the same need for each. A finance lead asks about risk and cost. A founder asks about speed and setup. An operations manager asks about fit with the tools they already use. Semrush advises mining support tickets, sales calls and forum posts for the exact wording 1. Our guide to finding the questions people ask ChatGPT lists more places to look.

Do not write ten near-identical variations of one question. They test the same thing and make your results look steadier than they are.

Should branded prompts be in the list?

Yes, but keep them in their own bucket and report them on their own line. Branded prompts name you, so you will usually appear, and they lift the overall rate. Semrush makes the same point: report branded and non-branded prompts separately 1.

The headline number to watch is your non-branded mention rate, meaning the share of non-branded prompts where the AI names you. Branded prompts answer a different question: is what the AI says about you true? Check them for wrong prices, old features and missing services. If you find errors, see how to fix wrong information AI says about your brand.

How many prompts do you need, and how often should you run them?

Start with 30 and grow only when a group gets noisy. With six prompts in a group, changing a single prompt can move that group's rate by about 17 points, so Semrush suggests expanding groups to 15 or more over time 1. Keep the core prompts stable for at least a month, because a trend line needs a constant list 1.

Frequency matters less than most teams fear. Profound compared one run a day with ten runs a day over 14 days in June 2026, using 753 prompts. Visibility came out at 78.7% against 80.4%, and the authors concluded that once a day is enough in almost every case 4. That is a vendor study of its own data, but it points the same way as common sense: a steady schedule beats a heavy one.

Paid tools cap the list. Otterly lists 15, 100 and 400 prompts on its three self-serve plans, at $29, $189 and $489 a month on the prices we read 7. A manual run of 30 prompts across three assistants is 90 checks, which is a long afternoon once a month. Plan the size around the time or money you will really spend.

What does a finished list look like?

Here is an invented example, not a real client. A 12-person bookkeeping firm in Leeds serves restaurants and wants new clients. Two buyer types: an owner who is overwhelmed, and a manager who inherited the books. The list has six prompts per bucket, and the five below show how the wording shifts. Every number in this example is made up for illustration.

Five of 30 prompts for a Leeds bookkeeping firm (illustrative)
  • Problem | owner | I'm six months behind on VAT for my restaurant, what should I do first?
  • Category | owner | Who are the best bookkeepers for restaurants in Leeds?
  • Comparison | manager | Local bookkeeper vs online bookkeeping app for a restaurant with 15 staff
  • Selection | manager | What should I ask a bookkeeper before hiring them for a restaurant?
  • Brand | owner | Is [Firm name] a good bookkeeper for restaurants, and what do they charge?

What mistakes make a prompt list useless?

Most failures come from a list that flatters you or keeps changing. Check these before your first run.

  • Too many branded prompts. The score looks healthy and tells you nothing about new buyers.
  • Keyword-style prompts such as “bookkeeper Leeds”. People talk to AI in full sentences.
  • Editing prompts mid-month. You then compare two different lists and call it a trend.
  • Chasing one prompt. One answer can change from run to run, so judge groups and months, not single results.
  • Writing separate pages for every prompt. Google treats separate pages for every query variation, made mainly to sway rankings or AI answers, as scaled content abuse 9; our keyword picking guide shows how to group them. Use the list to measure, not to multiply pages.

When is a prompt list not worth building?

Skip it if your site has not yet been indexed or you have fewer than 5 pages that answer a buyer question. You would be measuring an empty room. In that case, run 5 prompts by hand once a month and spend the rest of the time publishing. Grow the list as your pages grow.

Frequently asked questions

How many prompts should I track for AI visibility?

Start with about 30, six in each of five intent buckets. Semrush's walkthrough starts with about 30, six in each of five intent buckets, which is small enough to run by hand 1. Add prompts only when a group's results swing wildly, and expand each group toward 15 over time.

Can I find out how often people really ask a prompt?

No. Ahrefs says no major AI platform except Microsoft Copilot shares its query data, so tool volumes are estimates built from Google search volume 5. Choose prompts from your own customer language and judge them by whether buyers would plausibly ask them.

Should I include my brand name in prompts?

Include a small branded group, then report it separately. Branded prompts show whether the AI describes you accurately, but they inflate your overall mention rate 1. Your headline figure should come from the non-branded prompts.

How often should I run the list?

Monthly by hand is enough for most small teams. Profound's June 2026 test found once a day came close to ten times a day for visibility 4, so running more often adds cost faster than insight. Keep the same schedule so months compare.

Is there a free way to generate the first draft?

Yes. The AI Prompt List Builder turns your brand and category into 30 to 50 prompts as a CSV. Treat its output as a draft, then replace generic lines with wording from your own customers.

About the author

Rahil Jain

Rahil Jain is an AI SEO (answer engine and generative engine optimisation) consultant and co-founder of Chalk Labs.

LinkedIn · Method and credentials

Sources

  1. How to Build an AI Visibility Dashboard in Semrush · SemrushOpened October 20, 2026. Supports: Five prompt groups, start with about 30 prompts (six per group), a one-prompt change moving a group by 17 points, report branded and non-branded separately, keep core prompts stable for a month, mine support tickets and sales calls.
  2. The Answer Economy: How AI Search is Rewiring B2B Software Buying (2026 AI Search Insight Report) · G2Opened October 20, 2026. Supports: Initial AI prompts of B2B software buyers: 33% category, 31% competitor, 22% requirements, 9% ecosystem, 6% budget (survey fielded March 2026).
  3. How ChatGPT's query fan-out works · AhrefsOpened October 20, 2026. Supports: 14.4% of 66.6 million logged-out ChatGPT responses contained fan-out queries, about 1.6 per response (August to September 2026).
  4. Is once a day enough? · ProfoundOpened October 20, 2026. Supports: Once-a-day versus ten-times-a-day tracking over 14 days in June 2026 on 753 prompts: visibility 78.7% vs 80.4%; authors conclude once a day is enough almost always.
  5. How Ahrefs estimates AI search demand · AhrefsOpened October 20, 2026. Supports: No major AI platform except Microsoft Copilot shares query data; AI adjusted volume is a proxy from Google volume, to be used directionally.
  6. AI features and your website · Google Search CentralOpened October 20, 2026. Supports: AI Overviews and AI Mode may use query fan-out across subtopics and data sources.
  7. Otterly.AI pricing · Otterly.AIOpened October 20, 2026. Supports: Lite $29 (15 prompts), Standard $189 (100 prompts), Premium $489 (400 prompts) per month; page also shows a second, lower price set.
  8. ChatGPT search insights from 17 months of clickstream data · SemrushOpened October 20, 2026. Supports: Web search enabled for 34.5% of ChatGPT queries in February 2026 in US clickstream data.
  9. Optimizing your website for generative AI features on Google Search · Google Search CentralOpened October 20, 2026. Supports: Creating separate content for every query variation primarily to manipulate rankings or generative AI responses violates the scaled content abuse policy (page updated 2026-07-10).
How this page was made. Built on Rahil Jain's AI SEO method and drafted with AI research assistance. Every statistic links to the page we read it on (opened October 20, 2026 or nearby). Published October 20, 2026; last updated October 20, 2026. Chalk Labs sells AI SEO services, so we have a commercial interest in this topic. Spotted an error? Tell us and we will fix it.