The one-page version
For a B2B firm that wants to stop guessing what to publish and start writing against questions its actual buyers ask.
- 1Start with your own sales calls and inbox. Every question a prospect asks you is a question someone else is asking an assistant.
- 2Ask the assistants directly. Run your category questions through ChatGPT and Perplexity, then read the follow-up questions they suggest.
- 3Read where your buyers talk: LinkedIn comments, trade forums, industry groups, supplier RFQ emails.
- 4Use search tools last, to size and validate, never to originate the list.
- 5Sort by commercial value: who asks it, how close they are to buying, and whether you can answer it better than anyone.
- 6Put the survivors in one question map with an owner, a target page and a date.
The short answer
To research B2B buyer questions for AI search, collect real questions from four sources: your sales conversations, the assistants themselves, the places your buyers talk publicly, and search data. Then score each question on who is asking, how close they are to buying, and whether you hold evidence nobody else has. Publish against the top of that list.
What you are building is not a keyword list. It is a map of the decisions your buyer has to make before they can hire you, written in their words, in full sentences, with the awkward ones left in.
Why AI search questions are not keywords
A keyword is a compressed guess at intent. "Linkedin ads agency" could be someone comparing suppliers, someone researching a job, or someone who wants to know what one costs. You write one page and hope it serves all three.
A question carries its own context. "How much does a LinkedIn ads agency charge per month for a small B2B firm in the UK" tells you the buyer, the stage, the market and the budget concern in one line. That is a brief. It is also close to what people genuinely type into an assistant, because an assistant does not punish you for typing properly.
The practical consequence: your research should preserve the whole sentence. Every time you shorten a real question into a keyword, you throw away the part that made it useful.
A keyword tells you what someone typed. A question tells you what they are worried about.
Your own sales calls and inbox
This is the source everyone has and almost nobody mines. If a prospect asked it on a call, they had already tried to answer it themselves first, and increasingly that first attempt is an assistant. Your call recordings are a log of the questions the market could not resolve on its own.
Go back through the last twenty calls and pull every question verbatim. Not paraphrased into marketing language. The awkward phrasing is the point: it is what the next person will type. Do the same with your inbox, your proposal follow-ups and the objections that killed deals.
Pay particular attention to the questions asked after you sent a proposal. Those are the highest-value questions on your list, because someone was ready to spend money and something stopped them.
One row per question. Ten minutes per call recording.
QUESTION (verbatim): ______________________
WHO ASKED: role, company size, sector
STAGE: first call / mid / post-proposal
WHY THEY ASKED IT: what worry sits underneath
WHAT I SAID: the answer that landed
EVIDENCE I HAVE: client data, screenshot, number
DO WE HAVE A PAGE FOR THIS? yes / no / partly
Post-proposal questions go straight to the top of the map. Someone had their wallet out and hesitated.
The assistants themselves
Ask ChatGPT and Perplexity your category questions and read two things: what they answer, and what they suggest you ask next. Those follow-up prompts are a decent proxy for how the model has learned people navigate the topic. They will surface angles you have never written about.
Then run the questions you already care about and note which sources get cited. If the same three sites keep coming up, read what they did structurally. Often it is nothing clever, just an actual direct answer in the first paragraph while everyone else warmed up for six.
Log this properly rather than doing it once in a browser tab you close. Date, prompt, assistant, whether you appeared, who did. Without the log you cannot tell in three months whether anything you published changed the picture.
Where your buyers actually talk
For most niche B2B markets this is not Reddit. It is LinkedIn comment threads under posts by people in the sector, a trade association forum, a WhatsApp group you are in, the questions asked at an industry event, or the specification queries that arrive with an RFQ.
Look for the questions that get asked repeatedly and answered badly. Repetition tells you the market has an unresolved need. Bad answers tell you the citation slot is open. That combination is the best signal in the whole exercise, and no tool will hand it to you.
One caveat worth stating: what people ask in public is a subset of what they ask in private. Nobody posts "am I about to waste twelve grand" on LinkedIn. That question still gets typed into an assistant at eleven at night, which is why source one matters more than this one.
Search data, used properly
Search tools come last, and they do a different job. They cannot tell you what people ask assistants, because that data is not published. What they can do is confirm a question theme exists at scale, show you the phrasing variants, and stop you writing three pages that all answer the same thing.
Your own Search Console is more useful here than any third-party tool. The long queries at the bottom of the impressions list, the ones with two impressions and no clicks, are frequently the exact full-sentence questions you should be answering. Sort by query length rather than volume and read the tail.
Do not filter your list by search volume. Many of the best B2B questions have no measurable volume at all, because only forty people in the country ask them and every one of them could be a client.
Zero search volume and forty possible buyers is a better page than ten thousand searches and none of them can hire you.
Want more of the right enquiries, not just more traffic?
Book a 30 minute call. We will look at where your leads come from now and what would actually move the number.
Sorting questions by commercial value
You will finish the collection stage with a couple of hundred questions and no idea where to start. Score each one out of three on three things, and write against the highest totals first.
Is it the person who signs, someone who influences them, or someone who will never buy? A question asked by students and job seekers scores zero however popular it is.
Pricing, comparison, 'how do I choose', 'what goes wrong' and 'is it worth it' sit near the money. Definitions sit a long way from it.
Do you hold evidence nobody else has? Client data, a real number, a mistake you made and fixed. If your answer would be interchangeable with everyone else's, skip it.
Nine out of nine is rare and you should drop everything to write it. Seven and above is your publishing queue. Below five, leave it: the page will be fine, nobody will care, and it will dilute the topical focus of the site.
Building the question map
One sheet. Every surviving question, grouped into clusters, with a target page, an owner and a date. The clustering matters more than it sounds: five closely related questions belong on one strong page with five headed sections, not five thin pages competing with each other.
Copy this into a sheet. It is the whole system.
QUESTION (verbatim, full sentence)
CLUSTER (the topic it belongs to)
SOURCE (call / assistant / forum / search console)
WHO ASKS /3 INTENT /3 CAN WE WIN /3 TOTAL /9
EVIDENCE WE HOLD: the number, client or screenshot
TARGET PAGE: new page / new section on existing page
OWNER + PUBLISH DATE
CITED? checked monthly, with the date
The last column is what turns this from a content plan into a feedback loop. Without it you never learn which questions were worth answering.
Review it monthly. Add the new questions from the month's calls, re-check the citation column, and move anything that has been sitting at the top of the queue unpublished for two months either into the calendar or out of the sheet.
The first 30 days
- Week 1Mine twenty sales conversations
Call recordings, proposal follow-ups, the objections that killed deals. Verbatim questions only, with the role and stage of whoever asked.
Output: 60 to 100 real questions, in your buyers' own words. - Week 2Run the assistants and read the tail of Search Console
Prompt ChatGPT and Perplexity across your category, log who gets cited, then sort Search Console queries by length and read the long ones nobody clicks.
Output: A citation baseline and 30 to 50 more questions. - Week 3Score and cluster
Three scores out of three on every question, group into clusters, decide which cluster becomes a page and which becomes a section on a page you already have.
Output: A ranked question map with the top ten clear. - Week 4Publish against the top question
One page, answering one cluster properly, with your own evidence in it. Then check monthly whether the assistants have started using you.
Output: First page live and a review date in the calendar.
Where this goes wrong
- Tidying up the phrasing. You rewrite "why is my LinkedIn spend going nowhere" into "LinkedIn ads performance optimisation" and lose the entire signal. Keep the sentence as it was said.
- Generating the list instead of collecting it. A model-written question list produces model-average content, which is the one thing that guarantees you are not the interesting source.
- Filtering by volume. The most commercially valuable B2B questions routinely show zero. Volume is a sizing input, never a gate.
- Building the map and never publishing. Research is the comfortable part. A half-decent map with four pages published beats a beautiful one with none.
- Answering questions you cannot answer better than anyone. If you have no evidence and no experience of it, you are writing a summary of other people's work. Nothing gets cited for that.
The question list is not the work. The evidence you can put behind the answer is the work.
Rather have someone run this alongside your campaigns?
Book a 30 minute call and we will talk through your pipeline, not your content calendar.
Read next
The five pillars that turn a question map into citations you can measure.
The 20-point audit: what on your site is retrievable, and what is unquotable.
The five-block structure that turns a brochure page into something worth quoting.
The four types a B2B site needs, the ones to skip, and how to validate them.
Common questions
If you want content and campaigns working together to bring in leads
- LinkedIn, Meta and Google campaigns run against a named target account list.
- Content and service pages built around the questions your buyers actually ask.
- Tracking and attribution wired up so you can see what is really driving enquiries.
- Creative built from your own material, shipped and tested every month.
- Reporting on cost per qualified pipeline rather than cost per click.
Starter £1,250/mo, Growth £1,725/mo, Scale £4,025+/mo. Content work sits inside the retainer where it is in scope. Anything outside that is quoted properly before it starts.