Rethinking SEO: Why Authority Isn't Everything for AI Citations

What changed in this AI citation study?

The change is not that authority stopped mattering. The change is that FactoryJet's analysis found authority did not predict citation frequency within its measured sample of Google AI Overview results. That distinction matters because organic ranking and selection as a supporting source in an AI answer are related outcomes, not interchangeable ones.

FactoryJet captured 232 Google AI Overview citation URLs across 12 commercial e-commerce queries. After excluding publisher, marketplace, and platform-vendor domains, it fetched and measured 58 of the most-cited pages using rendered HTML. The study reports a median cited-page length of 2,813 words and a median of 110 list items. It also found that 25 of the 58 pages were between 2,000 and 3,500 rendered words.

The authority result is the part most likely to unsettle an SEO playbook. In a separate comparison of agencies named by ChatGPT and Perplexity across nine buyer prompts, netalico.com had 557 referring domains and 17 LLM mentions. outerboxdesign.com had 4,718 referring domains and was described as barely present on those prompts. That is a before-and-after change in emphasis: before, many teams treated authority as a proxy for whether AI would surface them; after this study, authority is a necessary context signal but not a sufficient explanation for citation presence.

The study is useful because it measures pages that were actually cited, rather than starting with a theory about which technical trick should work. It is not proof that word count, lists, or a visible FAQ causes a citation. FactoryJet explicitly describes the results as correlation, not causation. That restraint is the correct way to read the data.

Google's own guidance supports the narrower interpretation. Google says that standard SEO best practices remain relevant for AI Overviews and AI Mode, that there are no additional requirements to appear, and that there is no special schema.org markup required. A page still needs to be indexed and eligible for a Search snippet. The opportunity is not a separate shortcut called AI SEO. It is stronger coverage that remains easy to retrieve, understand, and support.

1. Before the study: authority was often treated as a broad proxy for AI visibility.
2. After the study: citation behavior must be measured directly, because authority did not track LLM mentions in this sample.
3. What remains true: indexability, ordinary SEO fundamentals, and reliable content still matter.

What did the cited pages have in common?

The cited pages were not uniformly huge, heavily marked up, or formatted as listicles. They were more often substantial, structured pages with many extractable units. FactoryJet found a median of 110 list items per cited page, while only 18% used a numbered-listicle headline. The implication is not that every page needs a listicle. It is that useful lists can make claims, steps, criteria, and comparisons easier to retrieve than a wall of prose.

Length clustered in the middle. The study measured one page below 500 rendered words, two from 500 to 1,000, 13 from 1,000 to 2,000, 25 from 2,000 to 3,500, 14 from 3,500 to 6,000, and three above 6,000. The largest group sat between 2,000 and 3,500 words. That does not create a universal word-count target, but it does argue against padding every page past 5,000 words for AI visibility.

Visible FAQs showed up more often than FAQPage markup. A visible FAQ section appeared on 72% of the cited pages, while FAQPage schema appeared on 41%. The useful conclusion is practical: answer the follow-up questions on the page when readers need them. Do not assume markup can replace the visible answer, and do not add a FAQ merely to chase a signal.

One H1 was the most consistent structural trait. FactoryJet found exactly one H1 on 55 of 58 cited pages, with none having zero. This is ordinary publishing hygiene, not a proprietary citation lever. It still reinforces a basic lesson: a page should make its main subject unambiguous to people, crawlers, and answer systems.

Freshness appeared often but was not universal. dateModified was present on 56% of the cited pages. Updating pages when the underlying answer changes is sensible. Changing a date without improving the answer is not a citation strategy. A date field is evidence of recency only when the content earns it.

1. Use lists when they make a decision, process, comparison, or set of criteria clearer.
2. Build complete pages before adding length. The study's middle-length cluster is evidence against filler.
3. Include a visible FAQ when the question has predictable follow-ups.
4. Keep one clear H1 and make the page's central answer obvious.
5. Update content when facts, products, or buyer questions have changed.

Who does this affect?

This affects B2B SaaS and tech growth teams that have spent years treating conventional authority metrics as the main scoreboard. Those metrics still explain part of how a site earns search visibility. They do not tell the team whether ChatGPT, Perplexity, Gemini, Copilot, or Google AI Overviews cite the brand when a buyer asks a category or comparison question.

It also affects local, multi-location, and service businesses. A local firm can have fewer referring domains than a national incumbent yet still create a direct, well-supported answer to questions such as which provider serves a location, what a service includes, or how two options differ. That does not eliminate the need for reputation and links. It means the company should not wait for an enterprise-scale link profile before publishing the answer its market needs.

Editorial teams are affected because their work is now closer to the retrieval unit. Thin category pages, vague service pages, and generic thought-leadership posts leave an AI system with little concrete material to cite. A strong page gives a clear answer, explains the conditions behind it, and includes the facts a reader needs to validate the answer.

SEO agencies are affected because reporting needs to separate ranking evidence from citation evidence. A rising Domain Rating or referring-domain count can be good news. It cannot stand in for a direct measurement of whether the company appears in relevant AI answers. That is the gap Citation Share is designed to expose: the percentage of relevant AI answers in a category that cite a brand.

The study should not be used to tell a client that backlinks no longer matter. FactoryJet itself says its finding is narrower than that. It examined a limited set of commercial e-commerce and agency queries, and its structural analysis covered Google AI Overviews rather than every answer engine. The responsible takeaway is to broaden the measurement model, not replace one oversimplification with another.

1. Growth teams need separate reporting for rankings and Citation Share.
2. Service businesses need direct pages for the questions buyers ask by service and location.
3. Editors need to produce answerable pages, not generic volume.
4. Agencies need to stop presenting authority metrics as proof of AI-answer visibility.

How should you respond without chasing a new tactic?

Respond by auditing the questions where being cited changes a buying decision, then improving the pages that can answer those questions with evidence. Start with category, comparison, implementation, pricing-context, and location-specific questions that already matter to your pipeline. The goal is not to manufacture a format. It is to make the best available answer easier to cite.

Measure the baseline first. Run a stable set of relevant prompts across the engines that matter to your audience and record the cited sources, answer presence, and competitor mentions. That establishes Citation Share rather than relying on a vague impression that a brand is visible in AI. Keep the prompt set consistent long enough to distinguish a content change from ordinary answer variation.

Then choose the weak pages. A page that ranks but does not get cited may lack a direct answer, an explanation of trade-offs, explicit criteria, current facts, or useful structure. A page that is cited but incomplete may be an opportunity to expand carefully. FactoryJet's findings indicate that visible lists and complete answers are worth testing because cited pages commonly had them. It does not justify copying a median word count into every brief.

Use comparison tables only when the reader has a genuine choice to make. A table can separate scope, fit, constraints, and evidence in a way that makes the answer easier to scan. It should not disguise an unsupported claim. Each cell still needs to be defensible, especially when the page compares competitors or makes a recommendation.

Keep the technical foundation clean. Google states that pages need to be indexed and eligible to appear with a Search snippet to be shown as supporting links in AI Overviews or AI Mode. It also says no new machine-readable files, AI text files, or special schema are needed. Make core content available as text, maintain sensible internal links, and ensure structured data matches visible content.

Finally, evaluate outcomes over time. Citation Count per day measures volume. Answer Presence measures how broadly a brand appears across the question universe. Share of Voice compares it with named competitors. Citation Share is the headline measure because it asks the hard question directly: in relevant AI answers, how often is the brand the source chosen to support the response?

1. Define a fixed set of commercial and informational buyer questions.
2. Record baseline citations and competitors by engine before changing pages.
3. Improve the answer, evidence, and structure on pages that matter most.
4. Preserve conventional SEO work, including indexability and internal linking.
5. Compare Citation Share, Answer Presence, Citation Count per day, and Share of Voice after changes.

What should teams not conclude from the study?

Teams should not conclude that a low-authority site can ignore authority altogether. The study did not establish that links are irrelevant, and it did not measure organic ranking as its outcome. It found a mismatch between referring-domain counts and LLM mentions in a defined sample. That is a reason to investigate, not a reason to discard a durable SEO program.

Teams should not conclude that 2,813 words, 110 list items, or a visible FAQ are magic thresholds. Those figures describe the median page in a small, filtered set of cited pages. They do not predict the correct answer length or structure for every query, industry, or engine. A concise product specification could be the right result for one question. A long comparison may be needed for another.

Teams should not conclude that Google AI Overview findings apply automatically to ChatGPT, Gemini, Perplexity, Copilot, or Claude. FactoryJet states that its structural analysis did not measure pages cited by ChatGPT or Claude. Each engine can surface different sources. Omnicite tracks citation share across engines because a single-engine result cannot is the whole answer market.

Teams should not turn the study into a license for artificial list stuffing. Lists work when each item contains a self-contained, useful point. Repeating variations of the same claim creates a worse reader experience and weaker evidence. The page must remain credible to the person deciding whether to trust it.

The sharpest conclusion is also the most modest one. Authority is not everything for AI citations. The best response is disciplined measurement and stronger content coverage, not a claim that anyone has found a way to game the models.

1. Do not abandon link earning or technical SEO.
2. Do not convert descriptive medians into mandatory content rules.
3. Do not generalize Google AI Overview findings to every engine.
4. Do not add empty lists, thin FAQs, or unsupported comparison claims.

Sources

FactoryJet captured 232 Google AI Overview citations, measured 58 cited pages, and reported the study's authority, word-count, list, FAQ, and limitation findings. FactoryJet, 2026-08-03

Google states that SEO best practices remain relevant for AI has and that there are no additional requirements or special optimizations needed for AI Overviews or AI Mode. Google Search Central, 2025-12-10

Google states that a page must be indexed and eligible to appear in Google Search with a snippet to be shown as a supporting link in AI Overviews or AI Mode. Google Search Central, 2025-12-10