How to Find the Questions Buyers Ask AI (and Which to Track)

Short answer

Start with the questions buyers already ask you: sales calls, support tickets, lost-deal notes and reviews. Add what people ask on Reddit and in Google's People Also Ask, then rewrite each one as a full sentence with the context a buyer would give, such as team size, budget or tools. Test them in the engines, keep 20 to 50, and don't change the set. Keyword tools miss most of these questions, because most AI prompts don't match a search keyword.

Key takeaways

For most of a 17-month Semrush study, between 65% and 85% of ChatGPT prompts couldn't be matched to any traditional search keyword.
Your own conversations are the richest source: sales calls, support tickets, win and loss notes and reviews contain the exact words buyers use.
Buyers add context, so write each question the way they would, with team size, budget, tools or region.
Prompt volume numbers are modelled estimates, so use them to rank questions, not to count them.

Why isn't keyword research enough?

Because people ask AI differently than they search. OpenAI's study of ChatGPT use, an NBER working paper from September 2025 based on about 1.1 million sampled conversations, found that 49% of messages are "Asking": the user is seeking information or clarification to inform a decision. The paper says the "Seeking Information" topic, which includes products, "appears to be a very close substitute for web search."

The prompts are also longer. Google said at I/O 2025 that early AI Mode testers were "asking queries that are two to three times the length of traditional searches." In a Semrush study of US clickstream data, ChatGPT prompts that triggered a web search grew from 4.7 words in January and February 2025 to 8.7 words a year later, while prompts that didn't trigger search fell from 24.9 to 13.5 words. For most of the study period, between 65% and 85% of prompts couldn't be matched to any keyword in Semrush's database of more than 27 billion. The share using traditional search language rose from 18.9% in October 2025 to 34.9% in February 2026. This is vendor panel data, and the gap is narrowing, but most prompts still don't read like search queries.

Where do the real questions come from?

Source	What you get	What to watch
Sales calls and demo notes	The words buyers use for their problem and the alternatives they name	Capture the first question a buyer asked, not the summary
Support tickets, onboarding forms, chat logs	Questions from people deciding or just decided	Add "why did you pick us" to your onboarding
Win and loss notes	Which competitors came up and why you won or lost	Look for the deal-breaking requirement
Reviews of you and competitors	The jobs, complaints and comparisons that become "alternatives to" and "best for" questions	Read the complaints about competitors
Reddit and community threads	Informal phrasing, such as "anyone using X for Y?"	Use the wording, not the thread
Google Search Console	Queries that already reach you. AI features are included in the Performance report under the "Web" search type	Google describes no separate AI filter
People Also Ask	Related questions Google surfaces around a topic	Treat it as a prompt for variants
The engines themselves	Ask a seed question and note the follow-ups and the brands named	Do it logged out, and note the engine

Google's documentation says that sites appearing in AI features such as AI Overviews and AI Mode "are included in the overall search traffic in Search Console", reported "within the 'Web' search type." That makes Search Console useful for spotting which queries already bring you impressions, but Google describes no way to tell which of them were AI answers.

What shapes do buyer questions take?

Most fall into four shapes, each with context added:

Best for: "best project management tool for a 12-person creative agency"
Alternatives: "alternatives to Asana for a team that finds it bloated"
Comparisons: "Trello vs ClickUp for client work"
Problems: "how do we stop missing handoffs between design and development"

The context is what makes a buying question. Team size, budget, existing tools, compliance needs, region and experience level change who gets recommended. Citepoint's AI Recommendation Index uses fixed sets of 20 questions per category in exactly this style, such as "What CRM works best inside Gmail and Google Workspace?" and "Which CRMs are GDPR compliant and can store data in the EU?"

How do you turn raw material into a question set?

Collect 100 or more raw questions from the sources above, using the buyer's words.
Group them by intent: best for, alternatives, comparisons, problems.
Rewrite each as a full question with the buyer's context included.
Merge duplicates. "Best CRM for a small agency" and "which CRM do agencies use" are one question.
Cut to 20 to 50, favouring questions close to a purchase.
Lock the set, so trends compare like with like.

The free Buyer Prompt Generator drafts a starting list for your category if you want one to edit.

Do you need prompt volume data?

It helps for ranking, but it isn't a count. Ahrefs explains that its "AI adjusted volume" takes the Google search volume of a prompt's parent keyword and multiplies it by an estimated usage ratio for each AI platform, with prompts sourced from People Also Ask questions. Semrush says that "individual prompts are often too specific and unique to measure directly", so it calculates volume at the topic level. Neither tells you how often a specific prompt is typed. Use volume to decide which questions to tackle first, and don't read it as a number of people.

What does a finished set look like?

Here is a short sample for a fictional invoicing app, one question per shape plus context:

"What's the best invoicing app for a two-person design studio that bills in euros?"
"Alternatives to Ledgerly for freelancers who need multi-currency invoices"
"Tallybook vs Billwise for a small agency"
"How do I stop chasing late invoice payments without sounding rude?"
"Which invoicing apps work with Xero and accept card payments?"

Mix these shapes across 20 to 50 questions, and make sure each one is something a buyer who doesn't know you would plausibly ask.

How do you decide which questions to track first?

Ask each one on each engine and write down which brands are named. The questions that matter first are the ones where a competitor is named and you aren't, and where the answer maps to a buying decision. Rank those by demand. Citepoint does this automatically, listing every buyer question where AI names a competitor, scored by demand. A spreadsheet works too.

What mistakes should you avoid?

Only branded questions. Prompts that contain your name show how AI describes you, not how buyers find you. Track them as a separate group.
Questions that contain the answer. "Why is [your product] the best invoicing tool" tests nothing.
Over-polishing the wording. When SparkToro's volunteers wrote 142 of their own prompts about choosing headphones, the prompts had a semantic similarity of 0.081, so they were very different. Yet Bose, Sony, Sennheiser and Apple appeared in 55% to 77% of the 994 responses. Cover the intents rather than perfecting phrasing.
Changing the set every week. It breaks your trend lines.
Ignoring your market. Ask in the country and language of your buyers.

Frequently asked questions

How many questions should I track?

Twenty is enough to see a trend in one category, and 50 covers most companies' main use cases. Keep the set fixed so periods can be compared.

Can I ask ChatGPT to suggest the questions?

As a brainstorm, yes, but it produces plausible questions, not observed ones. Check each against real sources such as sales calls, reviews and Search Console, and keep the wording buyers actually use.

Should I include my brand name in the questions?

Keep two groups. Unbranded questions show whether buyers who don't know you are told about you. Branded questions show how AI describes you. Report them separately so one doesn't hide the other.

Does Search Console show which queries came from AI answers?

No. Google's documentation says AI Overviews and AI Mode appearances are reported in the Performance report under the Web search type, with no separate AI filter described. Use it to find queries that already reach you, not to isolate AI.

How often should I refresh the question set?

Add new questions when your product or market changes, and review the set about once a year. Avoid weekly edits, because every change makes visibility rise or fall for reasons unrelated to your brand.