Should our content answer the questions our customers actually ask?

Your sales team answers the same twenty questions every week. How long does it take. What does it cost. What happens if it goes wrong. Why should we use you rather than the firm down the road. They answer them well, from experience, and the answers win work. Then you look at your website and none of those questions are on it. The website talks about your values, your approach and your commitment to excellence.

That gap costs more now than it did two years ago, because buyers put those same questions to an AI assistant before they pick up the phone. In G2's April 2026 survey of 1,076 software buyers, 51% said they now start their research in an AI chatbot, up from 29% a year earlier. An assistant answering "how long does a factory fit-out take for a food manufacturer" looks for a page that answers that question. If your page talks about excellence and a competitor's page answers the question, the competitor gets named.

So yes, your content should answer the questions customers actually ask, in their words. This guide covers where to find them, which ones get firms named, how to write answers an assistant can quote, and a plan for the first three months.

#Why do AI assistants favour content built around questions?

Because a question is what they are given and an answer is what they have to produce. They reach for sources that already look like an answer to the question in front of them.

When a buyer asks ChatGPT "which accountants specialise in dental practices in the Midlands", a page headed "Accountants for dental practices: what to look for and who we work with" fits the shape of the answer. A page headed "Our services" does not. Where the answer sits matters too: an analysis of 1.2 million ChatGPT answers by Kevin Indig, reported by Search Engine Land in February 2026, found 44% of citations came from the first 30% of a page.

Questions also come in families. "What is a managed service agreement", "what should be in one", "how much should one cost" and "what happens when one ends" belong together. Google's AI Mode and AI Overviews use what Google calls query fan-out, running several related searches for one question (Google Search Central, July 2026), so a site that answers the whole family has more chances to be picked up. How to organise those pages is covered in how to organise content so AI assistants see you as the expert.

#How is a buyer's prompt different from a Google search?

A search is a few words. A prompt is a paragraph with context, several questions at once and a follow-up. Google says questions in AI Mode have been two to three times longer than traditional searches since it reached the UK in July 2025.

Example: the same need, written three ways.

A page that only says what you do answers the search. The prompt and the follow-up need pages on who it suits, how to choose, what goes wrong and what it costs.

#How do we find out what questions buyers ask ChatGPT about our industry?

OpenAI does not publish what people ask, so no tool can show you the prompts. You can get close by combining what buyers tell you with the public signals of what they search. Start inside your own firm, because you have better data than any keyword tool.

#Your sales team and call recordings

Ask each salesperson for the ten questions they hear most on first calls and the five objections at proposal stage, in the buyer's words. "Will this disrupt our production line" is a buyer's question; "implementation concerns" is not. If you record calls with a tool such as Gong or Fathom, search the transcripts for question marks and for phrases like "how do you" and "what happens if". And ask new clients directly: "what did you ask ChatGPT or Google before you called us?"

#Lost deals and tenders

CRM notes on lost deals show the questions you answered badly or too late. Tender clarification questions are gold: they are written by buyers, in their words, about exactly what worries them.

#Support inbox, onboarding and account reviews

Questions that arrive after the sale, such as "can we add another site", show what nobody explained, and prospects ask them too. Tag a month of tickets and the patterns show.

#Search data and the questions Google shows

Filter Google Search Console for queries starting with how, what, why, which, can, should or does. Look at Google's "People also ask" boxes for your main subjects, and use a tool such as AlsoAsked, which maps those boxes into families of related questions.

#Ask the assistants themselves

Put a buyer question to ChatGPT, Gemini, Claude, Copilot and Perplexity, and to Google AI Mode, and note the follow-up questions they suggest and which firms they name. That shows you who owns the answer today. For a routine version, see how to find out whether ChatGPT is recommending our firm.

Put everything in one list. In our experience most firms end up with sixty to a hundred questions. Group them by subject and by type.

#Which questions matter most?

Recommendation and comparison questions first, because those are the answers in which assistants name firms. Cost and process questions next, how-to questions after that, and plain definitions last.

A plain definition such as "what is a fractional finance director" is often answered from the model's own memory, without naming or citing anyone. A recommendation question such as "which fractional finance directors work with SaaS firms in the UK" has to name someone. Peec AI's study of 30 million sources, reported by Search Engine Land in March 2026, found review platforms such as G2 appeared often in recommendation queries, which is why reviews and third-party lists matter alongside your own pages.

The "names firms" column is our experience from running buyer questions for clients, not a published study. Definitions still matter as building blocks: put a short, clear one inside the pages that need it, or in a glossary that AI assistants can use as a reference.

For example, a commercial insurance broker's first pages might be "professional indemnity versus public liability for consultants", "which brokers specialise in cover for design firms", "how much professional indemnity cover does a design firm need" and "what happens if we under-insure", rather than "our approach to client service".

#Should we publish a "best [service] firms in the UK" page on our own website?

We would not publish a ranking with your own firm at number one. Buyers see through it, and assistants draw heavily on what others say about you. Publish fair comparison pages instead, and work to appear on lists and review sites that other people run.

A fair comparison page names the alternatives, says who each suits, and admits when a competitor is the better fit. It answers the question a buyer is asking when weighing options, which is why it gets used. For Helm Club, a founder membership community, "We built the structured data AI engines read, then a comparison page for every question a founder asks when weighing up peer groups: Helm against YPO, Vistage and the rest." The result: "Every comparison page we have published ranks first for its question on Google and the AI engines. The pages not yet built are the ones where a competitor still writes the answer."

#How do we structure an answer so an assistant can use it?

Answer first, reasoning second, example third, on every page and in every section.

The title should be the question in the buyer's words. The first paragraph should answer it in one or two plain sentences that make sense lifted out and read alone. Then explain why, then give an example a reader can picture. For ID-Pal, the rebuilt pages followed exactly this shape: "Five industry pages rebuilt: fund administration, wealth management, legal, accountancy and estate agency. Each opens with the question a compliance lead types, names the regulator, and answers inside the first ninety words." We take apart a page that does this well in what an article that ChatGPT actually quotes looks like.

A weak page opens with "Understanding professional indemnity", followed by a paragraph saying the subject is important and many firms find it confusing. Nothing has been answered. A strong page opens with "What does professional indemnity insurance cover?", followed by "Professional indemnity insurance covers the cost of claims made against your firm for advice, designs or services that caused a client financial loss. It pays legal costs and any compensation awarded, up to the limit you choose." Then the explanation, the exceptions and an example.

Be specific: "it covers claims for negligent advice, including legal defence costs" beats "it protects your business". Cover the different ways buyers ask the same thing through your headings and FAQ answers, written naturally. Do not add lines listing variant phrasings; that reads as the keyword stuffing described in whether ChatGPT and Google will quote content written for SEO.

#Do FAQ sections still help now Google has dropped FAQ rich results?

Yes, for readers and for follow-up questions, but not because of the markup. Google limited FAQ rich results to well-known government and health sites in August 2023 and stopped showing them altogether on 7 May 2026. FAQPage markup is still valid; it simply produces nothing visible in Google.

Nor is an FAQ a shortcut to ChatGPT. SE Ranking's November 2025 study, reported by Search Engine Journal, found pages with FAQ sections averaged 3.8 ChatGPT citations against 4.1 for pages without, probably because FAQs sit on simpler support pages. Write a short FAQ where buyers have real follow-up questions, answer each one directly, and keep any markup accurate without expecting it to rank you.

#How should we plan the first three months?

One good answer page a week, in a deliberate order, beats four thin posts a week in a random order. At that pace, three months gives about 13 pages on one subject, linked to each other.

Link each new page to the ones before it and to the service page it supports. Where existing posts already hold the answers, rewrite them to this shape rather than starting again. For how much output a firm like yours needs over a year, see how often we need to publish for AI assistants to notice us.

#How do we tell whether it is working?

Keep a fixed list of 20 to 50 of your buyers' questions, and ask each one several times a month in ChatGPT, Gemini, Claude, Copilot and Perplexity and in Google's AI features, in a clean session. Answers vary from run to run, so count a win only when it repeats. Record whether you are named, which page is cited and who else is named.

The questions where you are not named are your next pages, and where a competitor is named, their page shows what a good answer looks like. Also ask new enquiries what they read before they called.

#What to do this month

1. Ask your sales team and account managers for the questions they hear most, in the buyer's words, and add tender clarification questions and lost-deal notes from the CRM.
2. Pull question-shaped queries from Search Console and the "People also ask" questions for your main subject.
3. Group the list by subject and type, and pick the one subject where you most want to be named.
4. Write the first recommendation or comparison page for that subject, with the question as the title and the answer in the first two sentences.
5. Run ten of your buyers' questions in the main assistants and record who is named, as a baseline.

#Frequently asked questions

#What prompts do B2B buyers use when asking ChatGPT to recommend a supplier?

Longer ones than you might expect. They usually describe the firm (size, sector, location), the trigger ("our payroll manager is retiring"), and ask for a shortlist with reasons, then follow up on specific features, price or risk. Write pages that answer each of those parts, not only the service name.

#How many answer pages do we need before ChatGPT starts naming us?

There is no fixed number, and one page on a subject rarely does it alone. In our experience a connected group on one subject, answering the recommendation, comparison and cost questions, is what starts to show. Assistants that search the web live can reflect a new page within weeks; answers from a model's memory change only when it is retrained, which takes months. For ID-Pal: "Where a page has been rebuilt, ID-Pal appears. Where it has not yet, it does not, which tells us exactly what to build next."

#What if a competitor already answers the question well?

Answer it better, or move on. If their page is thin, general or out of date, a more specific answer with a real example and a named author gives an assistant a better source. If their page is excellent and the question is minor, pick a question nobody has answered well.

#How do we keep answer pages current?

Review the pages that bring in work every quarter and the rest at least yearly, update figures and examples, and show a visible updated date. Cost pages need the most frequent checks. How old is too old for AI assistants sets out the cycle.

#What if the questions buyers ask are not the ones we want to answer?

Answer them anyway. If buyers keep asking whether they can do it themselves, write the page that explains when they can and when they should not. That page is trusted because it is not a sales pitch, and skipping the awkward questions leaves them to a competitor.

The questions are already in your building. The work is writing them down and answering them where an assistant can read them. To see which questions your buyers ask and who is named today, start with a free Growth Review: 30 minutes with David Adams and three fixes in writing. The rest of our content strategy and authority guides pick up from there.

#Sources