The page types AI answers cite most, and how to build each one for your business

Your homepage is the page you polish most and the page AI answers cite least. In Wix's March 2026 study of 1,056,727 citations across ChatGPT, Google AI Mode and Perplexity, homepages took 5.26% of them. Ranked list pages took 21.88%. Articles that answer one question took 16.68%. Product pages took 13.66%.

So I think the question most small businesses are asking is the wrong one. It is not "is my site good enough to be cited". It is "have I built the page types the engines actually pull from", and most sites I look at have built one of them, if that.

This guide is a ranked list of those page types, with the number behind each one and where it came from, and then a build for each one that a one-person business can actually do. Every figure here comes from a study I opened, with the date and the link. Where a tool refused me, I say so rather than make a number up.

One honest line before the list. I am building CiteScore, a tool that measures whether AI answers name you, so I read these studies for a living now. The method at the end is the free version of what it does.

What you'll have when you're done

• A ranked list of the page types AI answers cite, with the percentage and the source for each
• One ranked list page, one single-question article and one product or service page rebuilt the way the citation data says
• A title and URL rule you can apply to every page in ten minutes
• A clear idea of what does not move the needle, so you stop spending on it
• A free three-run spreadsheet that tells you in four weeks whether any of it worked
• [SCREENSHOT: the page-type table from the Wix study next to the tracking sheet with week-one and week-four runs]

Before you start

• Your own site, with the ability to add pages and edit titles and URLs. Any CMS works.
• Free accounts on ChatGPT and Perplexity, and a browser for Google. The checking method needs nothing paid.
• A spreadsheet. Google Sheets is fine.
• A day for the first two pages. The rest can take a month and should.

The ranking, and where each number comes from

Four studies did most of the work here, and I'd rather you knew what each one measured before you trust a number from it.

Wix AI Search Lab, 23 March 2026. 75,000 AI answers, 1,056,727 citations, three engines (ChatGPT, Google AI Mode, Perplexity), every cited page classified into one of eleven content types. This is the only study I found that ranks page types directly, so it anchors the list.

Evertune, April 2026. Over 40,000 URLs that ChatGPT cited heavily across about 50 product categories over 60 days. It found 50% of those citations were listicles, and 58% of the listicles were ranked lists.

Ahrefs, 15 April 2026. 1.4 million ChatGPT prompts, looking at why a retrieved page got cited or did not. This is the "how to build it" study.

Semrush, 21 July 2025. 5,000 keywords, over 150,000 citations, four platforms. Its line about subpages beating homepages is the one I'd put on a wall.

Here is the ranking from the Wix data, which I have rounded to one decimal.

Two things jump out at me from that table. The homepage, the page every business spends the most on, is seventh. And the comparison page, the page every SEO deck tells you to write, is eighth. The engines are not rewarding the pages we were taught to build.

The rest of this guide is a build for each of the top four, then a note on the fifth, then what to stop doing.

Step 1: Build the ranked list page

This is the page type that wins the money questions. "Best accountants for a small e-commerce shop in Leeds." "Best CRM for a two-person agency." When Wix sorted citations by intent, listicles took 40.86% of the commercial ones, nearly double any other type. Evertune's read of ChatGPT was blunter: half of what it cited heavily was a list, and 58% of those lists were ranked.

A small business usually refuses to write this page, because it means naming competitors. I'd write it anyway, for two reasons. The engine is going to answer that question whether you wrote the page or not, and right now it is answering it from a page somebody else wrote. And a list that includes you and four honest alternatives is more likely to be cited than a page that only mentions you, because the question was plural.

The shape that the data rewards:

• A title that is the question. "The 7 best [service] for [specific customer] in 2026." Ahrefs measured the similarity between the prompt and the cited page's title at 0.602, against 0.484 for retrieved pages that were not cited. The title is doing most of the matching.
• A ranked, numbered list, each item with the same three fields: who it is for, what it costs, and the one thing it does that the others do not. Ranked beats unranked in Evertune's data by a wide margin.
• Yourself in the list, honestly placed. If you are the best choice for one kind of customer and not another, say so. An engine quoting that sentence is still naming you.
• A date in the title and a last-updated line. Ahrefs found cited URLs averaged 1,064 days old against 1,432 for organic results, 25.7% fresher, and in its ChatGPT study the median cited page was about 500 days old. Fresh is not "this week". It is "updated this year".

Title:  The [N] best [service] for [specific customer] in [year]
URL:    /best-[service]-for-[customer]
Intro:  One paragraph. Who this list is for and how you picked.
List:   Numbered. For each: who it suits · price · the one difference
You:    In the list, with the same three fields, at the rank you deserve
Footer: Last updated [month year]. How often you re-check it.

Step 2: Build the single-question article

Second overall at 16.7%, and the clear winner on informational questions at 45.48% of citations, 2.7 times the average. This is the "what is", "how much does", "do I need" page.

The build rule here comes from the Ahrefs ChatGPT study, and I'd say it is the single most useful sentence in all this research. Cited pages had titles that matched the question. Pages with a natural-language URL slug were cited 89.78% of the time when retrieved, against 81.11% for pages without one. So the URL is part of the title now.

One question per page. Not a topic. A question.

• Title: the question as a customer types it, with one qualifier. "How much does a will cost in the UK in 2026" beats "Will writing: our guide".
• URL: the question in words. /how-much-does-a-will-cost not /services/wills/pricing-info.
• First two sentences answer it, with the number, the range or the yes/no. An engine quoting your opening is the whole point.
• Then the detail, in short sections with headings that are themselves questions.
• One fact the engine cannot get elsewhere. Your own price. Your own turnaround. A number from your own records. That is what makes your version of a common page the one worth citing.

Q: [the question, exactly as typed]
A (sentence 1): the answer with the number or the range
A (sentence 2): the one condition that changes it
H2: Why it costs that
H2: What changes the price
H2: What we charge, and what is included
Last updated: [month year]

Step 3: Rebuild the product or service page with the facts in text

Third overall at 13.7%, and first on transactional questions at 24.88% of citations. This is the page someone lands on when they are ready to buy, and the engines know it.

Most service pages fail here in the same way. The price is in an image, a PDF or a "request a quote" form. The hours are in a widget. The service area is implied. The engine cannot quote what is not in the HTML as text, and the Semrush study's finding applies directly: the engines cite specific subpages and help articles that match the question's detail, not pillar pages that gesture at it.

So the rebuild is mostly subtraction and plain text.

• One page per product or service, not one page for all of them. A category page is a different page type and it is next.
• Price, range or starting price as text, with what it includes and what costs extra.
• The constraints as text: turnaround, area served, minimum order, what you do not do.
• The two or three questions people ask before buying, each with a short answer, on the page itself rather than in a separate FAQ.
• Last updated, because a price from 2023 is a page the engine will skip.

Step 4: Give the category page a job

Fourth at 11.3%, and it earns its place on navigational and transactional questions, 18.31% and 14.97%. This is your "services" or "shop" page, and on most small sites it is a grid of boxes with no words.

The fix is not to make it long. It is to make it describe the set. One paragraph at the top that says who the category is for and how the items differ, then the list of items with one line each. The engine uses this page when the question is "what kinds of X does this company do", and a grid of icons cannot answer that.

Step 5: The pages you do not host

Discussion threads were fifth overall at 7.5%, and on Perplexity they were 17.35% of everything it cited, more than double the average. Profound's February 2026 read of about 700,000 ChatGPT citations to social platforms found 99% of its Reddit citations went to individual threads, not subreddits or profiles, 85.4% of its YouTube citations went to specific videos, and its LinkedIn citations split 25.4% personal profiles, 25.1% posts and 18.4% company pages.

These are page types. You just do not own the domain.

What that means in practice for a small business:

• Answer the real question in one thread where your customers already ask it, with the same detail you would put on your own article page. Link to your page at the end, not the start.
• One video that answers one question, titled as the question. Ahrefs' 75,000-brand study found YouTube mentions correlate with AI visibility at 0.737 for ChatGPT, higher than branded web mentions at 0.664 and far higher than Domain Rating at 0.266. I did not expect that, and I'd treat it as the cheapest lever on this list.
• A LinkedIn post that is the article, not a link to it, if your customers are there.

What not to spend on

Three things the data says to stop doing, or at least stop expecting citations from.

Schema markup, on its own. Ahrefs tracked 1,885 pages that added JSON-LD schema between August 2025 and March 2026 against 4,000 control pages. AI Overview citations fell 4.6%, AI Mode rose 2.4%, ChatGPT rose 2.2%, and the two rises were statistically indistinguishable from zero. Keep schema for the other things it does. Do not build a page around it.

The comparison page. 2.2% of citations overall. I know every SEO template has "X vs Y" in it. The engines are answering comparison questions from ranked lists instead, and a ranked list that includes both X and Y does the job of five comparison pages.

Ranking on Google as the whole plan. It matters a lot for one engine and much less for the others. Ahrefs found 76.1% of pages cited by Google AI Overviews rank in Google's top 10. Across six AI platforms, only 12% of cited URLs ranked in the top 10 for the original prompt, 28.6% on Perplexity and about 8% on ChatGPT. Semrush found most ChatGPT citations in its sample came from URLs ranking beyond position 21. So Google rank is the AI Overviews lever. For ChatGPT and Perplexity, the page type and the title are the lever.

Step 6: Check whether it worked, three runs at a time

Here is where I'd be careful, because one check is not a measurement. Ask ChatGPT the same question twice and you will often get two different source lists. I wrote a whole guide on that, with the research behind it, and the method is free.

The short version. Ten to fifteen buying questions, written the way customers type them. Three engines. Three runs per question per engine, each in a fresh chat, each week. Log the URLs that come back. Read the change from week to week, never the single snapshot. By week four you will know whether the ranked list page is being cited, and on which engine, and you will have spent nothing.

I did try the paid route for this guide. Ahrefs has a Brand Radar report of cited pages, and I called it for jordanhongtai.com through its API on 5 October 2026. It returned "Insufficient plan". So there is no Ahrefs citation data in this guide, and I would rather tell you that than paste a number I did not get.

What I'd do differently

Write the ranked list first, even though it names competitors. It is the page type that wins the commercial question, and the engine is answering that question from someone else's list today.

Make every title a question and every URL a sentence. That is the cheapest change on this list and the one with the clearest number behind it.

Put the price in text. The product page is third for a reason, and most of them are unreadable to the thing that would cite them.

Make one video. I'd say the 0.737 correlation surprised everyone who read that study, me included.

Stop expecting the homepage to do this job. It is the front door. The engines cite the rooms.