Which Content Formats Do AI Search Engines Cite Most?

The content formats AI search engines cite most depend on the question: "best X" blog lists took 43.8% of source links in Ahrefs' study of 750 ChatGPT prompts, published December 4, 2025. Those prompts asked about software, products and agencies. A service brand deciding whether to commission another blog post needs to check that sample before turning the finding into a content plan.

I would start with the customer's decision, then choose the format that can answer it with evidence. A comparison, an explanation and a page for booking a local service do different jobs. A citation share from one set of questions cannot settle all three. If you need a starting question, the visibility grader can help identify one to investigate.

Last verified: October 16, 2026.

What the listicle finding actually measures

The Ahrefs result describes the pages ChatGPT sourced for that selected prompt set. It includes first-party and third-party links. It does not measure every published listicle, so 43.8% is not the chance that your next list will be cited. It also does not establish that changing a page into a list causes more citations.

That denominator changes the decision. A research team asking comparison questions has deliberately created situations in which a list can be useful. You can learn from the observed source mix without treating the format as a ranking instruction.

Before using a study to assign work, read its prompt examples, engine, collection period, URL classification and denominator. "Share of cited URLs," "share of answers containing a citation" and "share of published pages that earned a citation" answer different questions. They should not share an unlabeled leaderboard.

Local hiring questions produce another page mix

In our published contractor-website analysis, blogs and resource pages received 1.5% of 74,969 citations. Homepages received 43.1%, while service, location and service-plus-city pages together received about 48%. The sample covered 32 home-services questions in 100 US and Canadian metros across four engines, August 31 through September 20, 2026.

That result has a narrower denominator too: citations to contractor websites. It excludes the directory, publisher and other third-party URLs that an engine may cite in the same answer. It cannot tell us that blog posts represent 1.5% of all local-search citations.

Nor can we subtract the Cheers percentage from the Ahrefs percentage and call the difference a format advantage. The questions, engines, dates, source universe and page labels differ. The useful lesson is to match the evidence to the decision you are making.

Our location-page guide covers the branch-page specification. The question here comes earlier: does the buyer need an explanation, a comparison, or a local provider they can contact?

Match the format to the decision

Consider a heat-pump company serving Charlotte. These illustrative questions are about the same service, but each reader needs something different before taking the next step.

The explanation needs accurate equipment information. The repair-or-replace guide needs criteria and a project-specific assessment. The service page needs current coverage and a clear contact path. Moving the same paragraphs into a numbered list would not supply any of that missing evidence.

Use the question to decide which existing page should improve. Our guide to how AI engines use different sources explains why the visible sources can also change by product.

Format cannot replace useful evidence

Google's current guide to generative AI features, read October 16, 2026, says there is no ideal page length and no requirement to break content into tiny pieces. It also says there is no special schema markup required for generative AI search. Relevant images and video can support a page when they help the reader.

I would treat those points as permission to make the page useful. A short answer can resolve a narrow question. A detailed comparison needs enough room for its criteria and exceptions. A demonstration can earn its place when seeing the process helps someone understand it. The format should follow that need.

Keep evidence and presentation separate when reviewing a draft. A tidy comparison with unsupported prices is still unreliable. A polished case study without permission to publish the customer's details is still unusable. A video with no explanation of what it demonstrates can leave the reader with the same question they arrived with.

Choose one page to improve this week

Use a small set of real buyer questions to choose one existing page for a substantive improvement. Keep the question, market and engine explicit when checking AI answers. Record which URLs were cited and whether the business was named; those are separate observations.

Then inspect whether the page actually serves that question. If the buyer needs a comparison, add supported criteria rather than more introductory prose. If the buyer needs to hire a local provider, make the service fit and contact path clear. If the buyer needs to understand a process, use an explanation or demonstration the business can stand behind.

Have the service or operations owner check the factual changes before release. Keep the original version and note what changed. After the page has been released and had time to be discovered, repeat the same checks. A later change in citations is an observation to investigate, with other changes and engine variability still in view. It is not proof that the new format caused it.

By the end of the week, the marketing owner should have one revised page, its supporting sources and a recorded check from the person responsible for that service.

Methodology and limits

This article interprets two previously published observational datasets; it does not run a new citation experiment or pool their results. Ahrefs examined 750 selected top-of-funnel ChatGPT prompts across software, products and agencies; its article is dated December 4, 2025. The Cheers Market Baseline figures classify 74,969 contractor-website citations from four engines, 32 home-services questions and 100 US and Canadian metros, August 31 to September 20, 2026. The Cheers figures are aggregate results. Full filters and URL-labeling limits are in the contractor-study methodology. Neither dataset measures the causal effect of rewriting a page into a different format.

Sources

• Ahrefs: Do Self-Promotional "Best" Lists Boost ChatGPT Visibility?, Glen Allsopp, December 4, 2025; read October 16, 2026. Supports the selected-prompt sample and the 43.8% source-link share.
• Google Search Central: Optimizing your website for generative AI features, page dated July 10, 2026; read October 16, 2026. Supports the guidance on page length, content chunking, markup and relevant visual media.
• Cheers: What Should Location Pages Include for AI Search?, analysis last verified September 23, 2026; source and method checked October 16, 2026. Supports the bounded contractor-website citation distribution.

Dylan Allen-Arnegård is the CEO & Co-Founder of Cheers, the local search platform for multi-location service businesses.