• AI visibility

Guide · AI search

What decides whether an AI engine cites my page over a competitor’s?

In the largest controlled test so far, four things decided which competing page an AI assistant cited first: topic match, list position, a stated price and a recent date. Completeness and trust signals helped less, and formatting barely mattered. The test was a simulation run by a software company, so read it as a strong signal, not a law.

By Underneath Team Guide Updated

The short version

  1. Four “gatekeepers” won in all six AI systems tested: topic match, list position, a stated price and a recent date, in a 252,000-trial simulation (opens in a new tab) by researchers at Sprinklr.
  2. Eleven of 18 content factors (61%) mattered in at least four of the six systems; layout and formatting changes did not.
  3. On live Google, position still leads: 41.7% of pages ranking 1 to 3 were cited by AI Overviews, against 20.1% at positions 7 to 10, in our study of 3,096 ranking pages.
  4. A review of 45 studies by Martinez (opens in a new tab) found relevance and position were the most reproducible levers, while generic rewrite tricks transferred poorly.

How did researchers test which page gets cited first?

They gave AI systems two near-identical pages that differed in one detail, then recorded which page was cited first. The study, by Vishwakarma and colleagues at Sprinklr, a customer-experience software company, is the cleanest head-to-head test published so far.

They started from 100 anonymized product review articles across 50 categories, such as consumer tech and fitness equipment. From these they built 1,440 scenarios. In each one, two versions of a page matched in facts, prices and length, but differed in exactly one of 18 factors.

Brand and publisher names were replaced with invented ones, so fame could not sway the result. The order of the two pages was swapped to cancel out any bias toward whichever came first. Six AI systems, including GPT-5.2, Gemini 2.5 Flash and Claude 3.5 Sonnet, ran 252,000 trials in total.

What are the four gatekeepers?

The gatekeepers are topic match, list position, a stated price and a recent date. All six AI systems agreed on these four, and the effects were so large that the weaker page was nearly shut out.

In the authors’ words, failing on any one “can eliminate citation odds regardless of other content strengths.” The topic test compared a page about the products asked about with one discussing unrelated products. The date test compared content dated 2026 versus 2019.

List position is different from the other three. It is the slot your page lands in when the engine gathers sources, and you cannot set it on the page itself. It comes from how well the page is found and ranked in the first place, which is still a search problem.

FactorWhat the better page hadWho controls it
Topic matchCovers the exact products asked aboutYour content team
List positionAppears first among the sourcesSearch ranking and retrieval
PriceStates the price plainlyYour content and pricing teams
Recent dateDated this year, not years agoYour content team

What matters once the basics are covered?

Completeness, trust and comparisons matter next, but less. Seven secondary factors helped in most systems once the four gatekeepers were met.

They fall into three groups. Completeness means listing specifications and covering the topic in depth.

Trust means confident wording instead of hedging, claims backed by evidence such as tests or certifications, and no internal contradictions. Competitive positioning means using the words the question uses and comparing the product with alternatives.

The systems also differed in how picky they were. Kimi K2 responded to 83% of the factors, while Claude 3.5 responded to 50% and Gemini 2.5 to 33%. The authors give a telling example of a weak page: a product description that ends with “Contact us for pricing details.”

Does formatting or tone change the outcome?

Barely: formatting changes had no consistent effect, and tone mattered in only some systems. Seven factors (39%) had weak or no effects across the six systems.

Turning a dense paragraph into organized sections did not move citation in a consistent way. A promotional tone, a weaker value proposition and weaker social proof mattered in only two or three of the six systems. The authors judged that too few to call a pattern.

Our own data on live Google points the same way.

AI Overviews are the AI summaries at the top of Google’s results. In our study of pages they cite, an HTML table added just +0.6 points to the chance of being cited. Organization schema, a tag that describes the company behind a site, added +0.5 points, which is no meaningful difference.

What should you do about it?

Fix the four gatekeepers before anything else, then work on completeness and trust. In practice:

  1. Match each important page to the exact question buyers ask, and answer it near the top.
  2. State prices on the page in plain numbers. In our pricing study, only 61.9% of plan prices quoted by AI assistants were fully faithful to the vendor’s page, so clear pricing also protects accuracy.
  3. Keep visible dates current, and only change a date when the content really changes.
  4. Treat ranking and retrieval as part of the job, since list position is a gatekeeper you cannot fix with copy.
  5. Then add specifications, comparisons and evidence for your claims, and remove hedging words.
  6. Do not spend a redesign budget on layout alone; the evidence does not support it.

If you want help turning this into a page-by-page plan, see our generative engine optimization service.

What does the research not tell us yet?

The strongest evidence comes from one simulated, vendor-run test, so several questions remain open.

  • Bigger source lists. The test used only two pages at a time. The authors note real systems often pull in “five to ten or more pages”, so crowded results are untested.
  • Real brands. Brands were anonymized on purpose. How much a famous name or trusted domain changes the outcome is unknown.
  • Other industries. The pages were consumer product reviews. Services, B2B software and local businesses were not tested.
  • Beyond the first citation. The study measured which page was cited first, not whether the brand was recommended, or whether anyone clicked.
  • Independence. The authors work at Sprinklr, the rewrites were generated by an AI system, and the method was piloted inside the company.

Frequently asked questions

Does putting prices on my website help AI assistants cite it?

In a simulation, yes: a stated price was one of four factors that decided the first citation in all six AI systems tested. That test used anonymized product reviews, so it has not been confirmed for live engines or for services.

Does page formatting affect AI citations?

The evidence says formatting alone does little. In the Sprinklr test, layout changes had no consistent effect, and in our Google study an HTML table added only +0.6 points to the chance of being cited.

Can a smaller site beat a bigger competitor in AI answers?

Possibly, though the research has only tested this in the lab. In the original study of Aggarwal and colleagues (opens in a new tab), as summarized by Martinez, adding citations helped the fifth source gain 115.1% in visibility while the first lost 30.3%.

Is a recent date enough to win the citation?

No. A recent date mattered in the simulation, but live results are mixed. In our freshness study, among dated pages on the same Google results page, a recent date made no difference to AI Overview citation (−4.1 points).

What is a gatekeeper factor in AI citation?

It is a condition that, if failed, can stop a page from being cited first regardless of its other strengths. The Sprinklr study found four: topic match, list position, price and a recent date.

Sources

Free strategy call

Some questions are easier to answer about your own business.

Bring the one that matters most. On a free 30-minute call we’ll take a first look at it and send you a short written read afterward.

Ask about my businessHow we work