Does your page answer what buyers ask AI? Six checks, one page, about 45 minutes

Cover graphic: how to get cited by AI assistants, six page checks on one page in about 45 minutes.

How do you get cited by AI assistants?

Short answer: To get cited by AI assistants like ChatGPT, Claude and Perplexity, give one page a sentence worth quoting for each question a buyer asks: what the product is, who it is for, what it costs, its limits, and how to buy. The assistants lift sentences, not pages, so a vague page gives them nothing to quote. Six checks on one page, about 45 minutes, and every fix is a text edit.

For readers short on time

  • An AI assistant cuts your page into passages and quotes the one sentence that answers the buyer’s question. If no sentence answers it, you are passed over.
  • This is the second half of getting cited: not whether the assistant can read your page, but whether the page, once read, answers the buyer.
  • Six checks grade that on one page: task coverage, explicit facts, first-party evidence, proof next to the claim, a real date, and question-shaped headings.
  • Start with the one page buyers land on from a search, usually a product or pricing page, not the home page.
  • No developer needed. Every fix is a text edit, and the first two checks carry half the points.

Every Thursday we send The Revenue Signal, our weekly newsletter: one study or how-to, with the numbers and the sources. Subscribe here.

Imagine your sales director forwards you a screenshot on a Thursday morning. A prospect asked ChatGPT which vendors in your category fit a company their size, and the answer named three competitors and quoted a pricing guide from a blog you have never heard of. Your AI is already answering for your company, and nobody has checked what it says.

The uncomfortable part is that your page was probably in the running. In our study of who AI assistants cite for field service software, the 108 answers carried 1,493 citations, and 1,181 of them pointed at comparison pages, pricing guides and other companies’ blogs rather than the websites of the 18 vendors we tracked. On the question about flat monthly fees against per-technician pricing, none of the 18 tracked vendors was named in 9 answers [1]. The answer is there, you just can’t see it: the assistants read vendor pages, but they quote the sentence that answers the buyer’s question, and most vendor pages do not contain one.

Getting cited has two halves. The first is access: can an assistant reach and read your page at all. The second, the one this guide is about, is whether the page, once read, answers the buyer. It is one slice of answer engine optimization, the work of getting your pages retrieved and quoted; we cover the wider picture in the B2B AI visibility gap. The checks below are the ones we group as Buyer Information and Evidence Quality: six checks worth 30 of the 100 points in our page-grading method, and every fix is a text edit on the page. No developer is needed unless your editor will not let you change body text. You will need one page of your website, a browser, and about 45 minutes.

 

 

Why is the best product in your category still passed over?

Two-column comparison: a vague line 'Plans for every budget' gives an assistant nothing to lift, while a concrete line 'from $49/mo, up to 25 staff' is a sentence it can quote and collects the citation.

An AI assistant does not read your page the way a visitor does. It cuts the page into passages of a few hundred words, finds the passage that best matches the buyer’s question, and lifts a sentence from it. If the buyer asked “what does this cost for a 40-person company” and your page says “Plans for every budget”, there is nothing to lift. The assistant moves on to a page that says “from $49 a month for up to 25 employees”, which is why pricing guides and comparison blogs collect so many citations [1].

Google says the same thing from its side. Its guidance for site owners on AI Overviews and AI Mode states: “There are no additional requirements to appear in AI Overviews or AI Mode, nor other special optimizations necessary”, and adds that you do not need new machine-readable files or markup to appear in them [3]. What it does ask for is the content itself: its self-assessment list opens with “Does the content provide original information, reporting, research, or analysis?” [4]. These six checks are that question, broken into parts a machine can grade.

One more thing to know before you start. The method first decides what kind of page it is looking at, from the address and the first 3,000 characters of text: commercial (product, pricing, service pages), editorial (guides and research), local-service (clinics, shops, branches) or informational (home and about pages). That profile decides which five buyer questions the page is graded against. This guide uses the commercial profile, because that is the page your buyers land on. Keep the sentences that carry your facts within the first 8,000 characters of visible text, about 1,200 to 1,400 words, because that is how far the grader reads.

 

 

What our buyers are asking this week

Three figures from the 6sense 2026 buyer report shown as counts: 81 of 100 buyers use an AI assistant in evaluation (was 94 of 100), 26 of 100 use one to discover vendors, 92 of 100 winners were on the initial shortlist.

The question in our corner of the market this week is whether buyers use AI assistants to find vendors at all, or only to tidy a list they already have. 6sense published its 2026 Global B2B Buyer Experience Report on 2 November 2026, built on a survey of about 4,600 B2B buyers. Its numbers, as counts: 81 of 100 buyers use an AI assistant somewhere in the evaluation, down from 94 of 100 a year earlier; 26 of 100 use one to discover vendors; most buying groups, 73 of 100, already hold two or more vendor preferences before they contact sales, what 6sense calls the initial shortlist; and the vendor that eventually won was on that initial shortlist in 92 of 100 journeys [2]. Read together: most buyers ask an assistant something during a purchase, about a quarter ask it who to consider, and the shortlist is close to final before anyone talks to sales. The page an assistant quotes when that quarter asks is the page that puts you on the initial shortlist.

A smaller signal points the same way for the facts themselves. One consultant gave three Claude models the same 20 Shopify search results and asked each to pick a product, 1,980 picks in all, with no web access. When the request named a budget and a skin type, 298 of 300 picks came in under the $25 budget, and the models explained their picks with the price and the wording on the listing. It is one person’s test of one company’s models, so treat it as a signal, not a study [5]. The signal is the one these checks grade: a stated price and a stated fit get picked; “premium quality” does not.

 

 

The six checks on one page, step by step

A six-row checklist to get cited by AI assistants: task coverage, explicit facts, first-party evidence with a date, proof next to the claim, a real date near the top, and headings that read like buyer questions.

Open the one page your buyers most often land on from a search: usually a product, service or pricing page, not the home page. Open it in a private browser window so you see what a stranger sees. Keep a sheet with six rows, one per check, and write down for each check the sentence you found or the words “not there”. That sheet is the brief for whoever edits the page.

 

Step 1. Check 2.1, Task coverage: the five buyer questions (15 minutes)

This is the heaviest check, 10 of the 30 points, and it grades five questions a buyer asks in order. For a commercial page they are: what is it, who is it for, what problem does it solve and with what result, what are the fit, limits and terms, and how do I buy or book. Each question earns 2 points for a plain quotable sentence, 1 point for a vague one, 0 for none. Read the page top to bottom and write down the sentence that answers each question. Not the headline that hints at it: the sentence.

The buyer’s question What earns 2 points An example that earns full credit
What is it? A definition in the first paragraph under the main heading, using the product’s proper name and a category word a buyer would search for Acme Payroll is a cloud payroll service for UK employers with 10 to 250 staff.
Who is it for? A sentence specific enough to exclude someone: role, company type, size band, situation, geography Built for finance managers at UK firms with 10 to 250 staff who run payroll in-house.
What problem does it solve, and what happens? The problem, the mechanism and a result with a sourced number, or the mechanism alone if you have no number Acme Payroll cut monthly payroll from 6 hours to 1.5 hours, the median in our 2025 survey of 62 customers.
What are the fit, limits and terms? A starting price or range, plan names, requirements, integrations and at least one honest limit Plans start at 49 GBP a month for up to 50 staff. It works with Xero. It does not run non-UK payroll.
How do I buy or book? A sentence next to the button that names the action and the mechanism Book a 20-minute demo with the calendar below, or email sales@acmepayroll.co.uk.
Table 1. The five buyer questions of check 2.1 for a product or service page, with an example sentence that earns full credit. The example company is made up. Source: our Buyer Information and Evidence Quality method.
Table 1 as an image, the same five rows as the table above: the five buyer questions of check 2.1 (what is it, who is it for, what problem it solves and what happens, the fit, limits and terms, how to buy or book), what earns 2 points for each, and an example sentence that earns full credit.

Tip: read your first paragraph aloud and ask whether a stranger who heard only that sentence in a chat answer would know what the product is. If the answer is “sort of”, it scores 1. Common mistake: the definition lives in the meta description or the hero image and never in the body text; copy it into the first paragraph. Second common mistake: “for businesses of all sizes”. Name your best-fit buyer; the one outside the band will keep reading anyway.

If you want to go one step deeper, write the 12 questions your buyers actually type before you grade the page against them. Our Academy Lab on buyer questions does exactly that, with a worksheet.

 

Step 2. Check 2.2, Explicit facts: five things stated outright (5 minutes)

Check 2.1 allows partial credit. Check 2.2 does not. It looks for five facts, 1 point each, that are stated in a sentence an assistant could copy as a fact: the offering named, the audience or service area, the scope or capability or hours, the terms, price, source or contact method, and the conversion route, date or business identity. A product name that appears only in the logo and the page title does not count, because an assistant cannot write “Acme offers X” unless a sentence says Acme offers X.

The test takes five minutes with the browser’s find box. Press Ctrl+F (Cmd+F on a Mac) and search the page for your product’s name. Count how many times it appears in a full sentence in the body, not in a heading, a button or an image. Do the same for your price (search for the currency sign), for your contact email (search for the @ sign) and for the word “for” followed by your buyer. One sentence that carries the fact is enough; it just has to be a sentence. What good looks like: “Acme Payroll Standard is a cloud payroll service for UK employers with 10 to 250 staff. Plans start at 49 GBP a month. Email sales@acmepayroll.co.uk to book a demo.” Three sentences, five facts.

 

Step 3. Check 2.3, First-party evidence: proof you can stand behind (10 minutes)

This check is worth 6 points and has three outcomes: strong (6), partial (3), absent (0). Strong means specific, first-party, attributable proof in the text: your own numbers with a date, a documented method, a named case with its outcome, product specifications. Partial means a real number with no source or date, or a named customer with no outcome. Absent means marketing claims only. Third-party statistics help the next check, but they are not evidence about your product.

Run the replacement routine on your page. Search the text for these words: leading, best, #1, world-class, seamless, trusted, thousands, millions, proven, cutting-edge, industry. For every hit ask what fact stands behind it. If there is one, write the fact with its date and source. If there is none, delete the phrase. Then look at every logo wall and award badge and add one sentence of text under it, because an image is not quotable.

Scores absent (0 of 6) Scores partial (3 of 6) Scores strong (6 of 6)
“The leading platform in the market, trusted by thousands.” “Over 400 companies use it. Customers tell us it saves hours every month.” “412 companies ran payroll on it in September 2026. In our 2025 customer survey (62 responses), the median customer cut monthly processing from 6 hours to 1.5 hours.”
A row of customer logos with no sentence A named customer with no outcome A named customer, the problem, what changed, and the date
“Experts agree” or “studies show” with no source A real number with no date or source A specification sheet in the body text: dimensions, limits, standards met, with the version date
Table 2. How check 2.3 reads the same page three ways. Examples built on our evidence-quality method; the companies and numbers in the examples are made up.
Table 2 as an image, the same three rows as the table above: how check 2.3 scores the same page three ways, absent (0 of 6), partial (3 of 6) and strong (6 of 6), across a superlative claim, a wall of customer logos, and an 'experts agree' line with no source.

A caution for companies under NDA, which includes us: you do not need a client’s name to score strong. A documented method, your own measured numbers on your own work, and specifications are all first-party evidence. What does not work is the adjective.

 

Step 4. Check 2.4, Claim and support: the proof sits next to the claim (5 minutes)

Check 2.3 asks whether evidence exists. Check 2.4, worth 4 points, asks whether it sits next to the claim it supports. The reason is the passage cut described earlier: a claim in your opening and its proof in a results section three screens down become two unrelated passages, one an unsupported assertion and one evidence with no claim. Strong (4 points) means each material claim has its number or named source in the same sentence or the next one; partial (2) means some do; absent (0) means claims everywhere and evidence nowhere.

The test: pick the three biggest claims on the page and, for each, measure the distance to its proof. Same sentence or the next: fine. Further than that: move the proof up, or move the claim down. When the proof is someone else’s data, name the source and year in the sentence and link the source name: “In 6sense’s 2026 report, 26 of 100 buyers used an AI assistant to discover vendors” is quotable; “research shows buyers use AI” is not.

 

Step 5. Check 2.5, Freshness: a real date near the top (5 minutes)

Two points, and the easiest to earn. Strong means a clear publication or last-updated date, a version number or an explicit time scope such as “prices effective from 1 January 2026”, visible on the page or in the page’s structured data. A bare copyright year in the footer earns 1; nothing earns 0. Assistants weigh recency so they do not repeat an old price or a retired feature, and an undated page cannot show that it is current.

Look at the top of the main content. If there is no full date, add one line under the main heading: “Last updated 12 November 2026”. In WordPress block themes you set this with a Date block switched to its “Modified Date” variation, and SEO plugins add the published and modified dates to the page’s schema. Two rules: change the date only when the content changes, and never use a script that prints today’s date on every load. Google lists changing a page’s date to look fresh, without a substantial change, as a warning sign of search-engine-first content [4].

 

Step 6. Check 2.6, Useful structure: headings that read like buyer questions (5 minutes)

Three points. Strong means the page is organised into sections, lists and tables that make answers easy to lift out; partial (1.5) means some headings over dense prose; absent (0) means a wall of text or a page built from image cards. A heading tells the assistant what the next passage is about, which is how the passage gets matched to the question. “Pricing” above a paragraph about prices is the difference between that paragraph being retrieved for “how much does it cost” and being missed.

The test: read only the headings of your page, top to bottom, as if they were a table of contents. Do they answer the five buyer questions in order (what it is, who it is for, how it works, pricing, integrations, how to start)? Slogans as headings (“Built different”) fail; plain nouns and questions pass. Then view the page source and search for the heading, list and table tags to confirm they are real elements rather than bold text or stacked columns; page builders often make a “pricing table” out of three columns of text, and the row and column relationship is lost. In the WordPress editor, a real heading comes from the Heading block, and a real table comes from the Table block with its header row switched on.

Put the answer in the first sentence under each heading, keep paragraphs to two to four sentences, and never begin a paragraph with “It also” or “As noted above”, because each passage has to stand alone when it is quoted without its neighbours.

 

What does a page that AI assistants cite look like?

Checklist of what a page that wins an AI answer contains: a first line naming product, buyer and category; a starting price; dated first-party evidence next to its claim; a real last-updated date; question-shaped headings; and one owner so prices never disagree.
  • A first paragraph that defines the product by name, names the buyer and names the category in the buyer’s words.
  • A visible “Pricing and requirements” or “Is it right for you?” section with a starting price, the plan names, the integrations and one honest limit.
  • At least one piece of your own evidence in the text, with a date, within the first 1,200 words, sitting in the same sentence as the claim it supports.
  • A “Last updated” line with a full date under the main heading, changed only when the content changes.
  • A heading outline that reads like the buyer’s five questions, built with real heading elements, and a real table wherever the content is a comparison.
  • One person who owns the page’s facts, so the price on the page and the price in the sales deck never disagree.

 

 

What mistakes keep a good page uncited?

The mistakes that keep a good page uncited: fixing the home page first, describing the company not the product, hiding price behind contact sales, proof on a separate case-studies page, treating the date as decoration, and trusting an automated score blindly.
  • Fixing the home page first. Buyers land on product, service and pricing pages from a search; the method grades one page at a time, so start where the buyer starts.
  • Describing the company instead of the product. On a product page, define the product and name the company once.
  • Hiding the price behind “contact sales”. A range or a starting point is quotable; “contact us” is not, and the competitor who states a price wins the comparison.
  • Putting the proof in a separate case-studies page. The grader reads only the page it was given; evidence elsewhere does nothing for this address.
  • Treating the date as decoration. A copyright year is not a date; a script that prints today’s date is worse than no date.
  • Trusting an automated score blindly. Any tool that grades a page can fall back to simple text rules when its AI grader is busy, and a good fallback score is not proof your sentences answer the buyer. Read the page yourself.

 

 

Your plan for this week

A plan in three windows: today, fill the six-row sheet on your main page in 45 minutes; this week, fix superlatives, move proof next to claims, add the date, fix headings; this month, repeat on the next three pages, then re-check the first.
  • Today: open your main product or service page in a private window and fill the six-row sheet (45 minutes). Write the five missing sentences from Table 1 yourself; nobody knows the product better.
  • This week: run the replacement routine on superlatives, move each proof next to its claim, add the date line and fix the heading outline. Hand the sheet to whoever edits the page.
  • This month: repeat the six checks on the next three pages buyers land on, then re-check the first page.
  • Do not spend on: new blog posts or a content retainer before the five buyer questions are answered on the pages you already have. New pages inherit the same gaps.

Short on time? Do Step 1 and Step 2 today. They take 20 minutes, need nobody else, and carry 15 of the 30 points in this set of checks.

 

 

How Revenue Experts helps you get cited by AI assistants

How Revenue Experts helps you get cited by AI assistants: the AI citation audit asks your buyers' 50 questions of ChatGPT, Claude and Perplexity, saves 450 answers over 3 runs, and for $1,495 in 2 to 5 days shows where you are named, who is cited instead, and what to fix.

Revenue Experts can do this for you. Our AI citation audit asks ChatGPT, Claude and Perplexity your buyers’ questions, saves every answer, and shows how often you are named, whose pages are cited instead, and what to fix, with the saved answers so you can repeat any check. 50 questions, 450 answers, 2 to 5 days, $1,495. To start, book a 20-minute call with John Bush, or tell us about your market on our contact page.

If this was useful, the next one arrives on Thursday. Subscribe to The Revenue Signal.

 

 

FAQ: getting cited by AI assistants

FAQ card: the six checks need no developer because every fix is a text edit, and a page can be graded 'informational' when its address and first 3,000 characters never say pricing or request a demo.

Do these six checks need a developer?

No. Every fix in this set of checks is a text edit on the page. You need a developer only if your editor will not let you change body text or add a real table.

Why was my pricing page graded as “informational”?

The profile is picked from the address and the first 3,000 characters. If the address is /plans and the text never says “pricing” or “request a demo”, the grader may pick the general profile. Add the plain words as a heading and a call to action, then grade it again.

We cannot name clients. Can we still score strong on evidence?

Yes. Your own measured numbers with a date, a documented method and product specifications are all first-party evidence. Names help; adjectives do not.

Will answering the five questions get us cited?

Nobody can promise that, and we do not. It removes the most common reason a page gives an assistant nothing to quote. What the assistants then say is a separate measurement, which we make with saved answers; we wrote up how we count it in how AI assistants decide which pages to cite.

How is this different from checking whether AI can reach my page?

Reaching and reading your page is a separate set of checks: can the assistant crawl it and render it. This guide is the next step, whether the page, once read, answers the buyer. A page can pass one and fail the other.


Related


Sources

  1. Revenue Experts, “Who do AI assistants cite when buyers ask which field service software to buy?” (18 vendors, 12 questions, 108 answers, 1,493 citations), 3 October 2026 (opened 14 November 2026).
  2. 6sense, “6sense Research Finds That Vendors Need to Be Known and Preferred Before the Buying Journey Begins” (6sense’s own announcement of its 2026 Global B2B Buyer Experience Report, 2 November 2026; opened 14 November 2026; states most buyers use an AI assistant somewhere in an evaluation, 81 in 100, down from 94 in 100 last year; only 26 in 100 use AI assistants to discover vendors; most buying groups, 73 in 100, already have two or more vendor preferences in place; and the eventual winner is on that initial shortlist 92 in 100 times).
  3. Google Search Central, “AI features and your website” (opened 14 November 2026; no change date shown on the page).
  4. Google Search Central, “Creating helpful, reliable, people-first content” (opened 14 November 2026; no change date shown on the page).
  5. Slobodan Manic (No Hacks), writing in Search Engine Journal, “3 AI Models Shopped Shopify’s Catalog. Each Picked A Different Winner”, 2 November 2026 (1,980 picks by three Claude models, no web access, one person’s test; opened 14 November 2026).

All sources opened and checked on 14 November 2026.