12 buyer questions to pull from closed-won deals before you research AI visibility
For B2B SaaS marketing, demand generation and RevOps teams who have won-deal data in the CRM but no list of questions to check in ChatGPT, Perplexity and Google's AI answers. Each of the twelve names the CRM field to look at, an example question to track, and what to do with the result.
The short answer
- A CRM does not hold the words buyers type into AI assistants. It holds who bought, how much they paid, where they came from and why deals were lost. Those facts tell you which questions are worth asking.
- Eight of the twelve questions below come from standard fields: job title, industry, company size, country, deal amount, win and loss counts, lost reason and original source. Four need a person to read deal notes.
- Turn each answer into two or three questions in the buyer's own words, ask them in each AI assistant several times, and write down who is named and which pages are cited.
- Sort the results three ways. AI cites you: keep that page fresh. AI cites someone else's page: ask to be listed there. AI cites nothing useful: write the page.
Why start from won deals
Most teams pick the questions to track from keyword tools or from a vendor's suggestions. That gives a list about the category in general. Won deals give a list about the buyers who pay you: their role, their company, the price they accepted and the objections they raised on the way.
6sense's 2025 Buyer Experience Report found that 94% of B2B buyers use AI chat tools during a purchase. The same kinds of people who are in your won deals are asking those tools about your category now. The work is to guess their questions well, and deal data makes the guess better.
What a CRM can and cannot tell you
It can tell you who buys. Group won deals by job title, industry, company size and country and a few groups will hold most of the revenue. It can tell you the price range that closes, and, if your team fills in a lost reason, what stops deals.
It cannot tell you the exact words anyone typed into ChatGPT. AI assistants do not share that. Every question you write from deal data is a model of what a buyer asks, so check it against sales calls and adjust.
Work with groups, not single records. A group of one or two contacts is one person's story, and it can identify them. Three or more is a pattern.
Questions 1 to 4: who buys
- 1. Which job titles are on the deals you win? Look at job title on contacts at won companies. Example question to track: "What should a RevOps lead at a 200-person SaaS company look for in contract software?" Buyers describe themselves when they ask AI, and the answer changes with the role.
- 2. How senior is the person who first got in touch? Look at seniority in the title. A manager researching asks "which tools do X". A director approving asks "is X worth the cost" and "what are the risks of X". Track both.
- 3. Which industry and company size bring the most won revenue? Cross industry with employee count. Example: "contract software for fintech companies with 200 to 500 staff". If one cell holds most of your revenue, most of your questions belong there.
- 4. Which countries do won deals come from? Look at country on the contact or company. Write the questions again in that market's language instead of translating them, because local buyers name local review sites, rules and competitors.
Questions 5 to 8: what they compare and what stops them
- 5. What does a won deal usually cost? Look at the middle of your won deal amounts and the range around it. Example: "How much does contract software cost for a 50-person legal team?" If AI quotes a price for you, check it is right.
- 6. Where is your win rate high and where is it low? Compare won and lost counts per segment. Where you win, track the questions to defend the position. Where you lose, track the questions to find out which pages make the case against you.
- 7. Why do you lose? Look at the closed-lost reason. Each common reason is a question buyers also ask AI: price becomes "cheaper alternatives to X", a missing integration becomes "does X work with Salesforce", a failed security review becomes "is X SOC 2 compliant".
- 8. Who do you lose to? This needs a competitor field or a read of the notes. Example: "X vs Y for mid-market SaaS" and "alternatives to Y". These comparison questions are where AI most often cites third-party roundups and review sites.
Questions 9 to 12: what triggers the search and what proof closes it
- 9. Where did won buyers first come from? Look at original source. Buyers who arrived from search asked category questions. Buyers who came by referral or direct already knew your name, so they ask AI "is X any good" and "X reviews".
- 10. What happened just before they started looking? Read the first-call notes of ten won deals. Triggers such as a failed audit, a new hire or an outgrown spreadsheet become questions like "we manage contracts in spreadsheets, when should we move to software?"
- 11. Which requirements came up in every won deal? Read the notes and security questionnaires. Integrations, data residency and certifications turn into yes or no questions, and AI answers them from your docs or from someone else's.
- 12. What proof closed the deal? Note which case study, reference call or review was shared before signature. Example: "X case study fintech" and "X customer reviews". If AI cannot find that proof on a public page, it cannot repeat it.
From questions to citation wins and next actions
A citation win is a buyer question where the AI answer cites one of your pages, or a page that names you. A citation gap is a question where it cites pages that leave you out. Ask every question several times in each assistant, because one answer is mostly luck, and record the pages cited each time.
Then each question falls into one of three groups, and each group has one action.
- AI cites your page: a win. Note which passage it used, keep the page current, and add the missing facts from questions 5, 11 and 12.
- AI cites a third-party page that lists tools like yours but not you: an outreach job. Ask the editor to add you, and give them the fact the page is quoted for.
- AI cites nothing that answers the question well, or only vendor home pages: a content job. Write one page that answers the question in its first two sentences and carries the proof.
The pages AI reads before it answersGet listed on the pages AI quotes
How Targetlytics does this from HubSpot
Connect HubSpot with read access and Targetlytics reads contacts, companies and deals. From contacts: job title, industry, country, lifecycle stage and original source. From companies: industry, employee count, annual revenue and country. From deals: amount, currency, stage, pipeline, won or lost, and the closed-lost reason when it is a picklist value. One sync reads up to 5,000 contacts, 5,000 companies and 2,000 deals, and it runs again every Monday.
Only counts are kept. Any group with fewer than three contacts is dropped, and no name, email address or company name is stored or shown to the language model that writes the proposals.
From those counts it proposes changes to your buyer types: up to six updates and up to three new ones per pass, plus brand facts such as use cases, pricing facts, competitors and category words. Each proposal shows the CRM group behind it with its contacts, won deals and revenue. You accept or dismiss each one. When you accept, the question list is rebuilt around those buyers.
The questions are then asked across AI assistants, the cited pages are listed, and the gaps arrive as a weekly to-do list: fix a page, write a page, or pitch a page that could list you. Outreach emails are drafted for you and sent only after you approve them. Questions tied to open deals move up the list.
Questions built from deals you have wonYour weekly to-do listHubSpot setup docs
What it does not do
- It does not read call notes, emails, deal names or free-text lost reasons. Questions 8, 10, 11 and 12 above still need a person, and you add what you find to your brand profile by hand.
- HubSpot is the CRM that connects today. Pipedrive, Attio and Zoho are coming. Salesforce is not on our integration list.
- The questions are our model of what buyers ask. No tool can see what real people type into an AI assistant.
- A fix or a listing does not guarantee AI will name you. We measure before and after so you can see what changed.
A note from the team that built the import
The mistake we see most is tracking the questions marketing would like buyers to ask. The CRM usually disagrees in two places. The company size that pays is larger or smaller than the one on the website, and the title that signs is not the title the ads target. Fix those two and the question list changes more than any keyword research would change it.
The second mistake is skipping lost deals. Lost reasons are the nearest thing a CRM has to a list of buyer doubts, and doubts are what people ask AI about late in a purchase. If your team leaves the field blank, making it a required picklist is the cheapest research you can do this quarter.
How to judge a tool for this job
- Does it read your own won and lost deals, or does it suggest questions from the category in general?
- Does every suggested question or buyer type show the CRM evidence behind it, with counts you can check?
- What leaves your CRM? Ask whether names and emails are stored, and whether small groups are withheld.
- Does it show the pages cited for each question, so you can tell a win from a gap?
- Does the gap come with a next step: a draft, a page to fix, an editor to contact?
- Do you approve changes and emails before anything is saved or sent?
How we wrote this
We make Targetlytics, so read this as a vendor's view. The product details were checked against our own code on 11 November 2026. The worked example below is made up to show the method; it is not a customer and its numbers are not results. When we have a real case we can publish, it will replace the example.
Worked example: from CRM evidence to a research plan
An invented 150-person B2B SaaS company that sells contract software. Illustrative only, not customer data. Scroll sideways on a phone.
| What the CRM showed | Question to track | What the AI answer cited | Group | Next action |
|---|---|---|---|---|
| Most won revenue sits in fintech companies with 201 to 500 staff. The website targets 51 to 200. | Contract software for a fintech company with 300 staff | Two review-site roundups; the company is in neither | Gap, third-party page | Pitch both roundups; change the buyer type to 201 to 500 |
| The title on won deals is Head of Legal Operations, not General Counsel. | What should a head of legal operations look for in contract software? | A competitor's buying guide | Gap, no neutral page | Write a buying guide for legal operations |
| The most common lost reason is "Missing integration". | Does [product] integrate with Salesforce? | The company's own integrations page, with an out-of-date list | Win, at risk | Update the page; say plainly what is and is not supported |
| The middle won deal is about €18,000 a year. | How much does contract software cost for a 50-person legal team? | A pricing comparison blog quoting an old price | Gap, wrong fact | Publish current pricing; ask the blog to correct it |
| Deal notes mention ISO 27001 in most won deals (read by hand). | Is [product] ISO 27001 certified? | Nothing about the company | Gap, no page | Publish a security page with the certificate date |
| Won buyers from referrals saw the same fintech case study before signing (read by hand). | [product] customer reviews fintech | The company's case study page | Win | Keep it current; add the numbers the buyer asked for |
Do it yourself in six steps
Pull won and lost deals
Export deals with their company and contacts, or connect the CRM. Include lost deals; the lost reason is half the value.
Group, then drop small groups
Count by job title, seniority, industry, company size, country, source and lost reason. Ignore any group with fewer than three contacts.
Answer the twelve questions
Eight come straight from the counts. For the other four, read the notes of ten won deals.
Write the buyer's questions
Two or three per answer, in the words the buyer would use, with their role and company size in the question.
Ask each question several times
In each AI assistant your buyers use. Record who is named and which pages are cited every time.
Sort and act
Win, third-party gap or content gap. Keep the page fresh, ask to be listed, or write the page. Ask again about two weeks after the change.
See which questions your won deals point to
Connect HubSpot and review the proposed buyer types and questions, each with the deal evidence behind it. Nothing changes until you accept it.
Measuring the traffic is half of it. The other half is knowing what to change next. See your to-do list.