TL;DR

The most rigorous public test of AEO claims — an Ahrefs study of 1,885 pages — found that adding schema markup produced no citation lift on Google AI Overviews or ChatGPT, and even a small decrease on AI Overviews for pages already cited. Google itself deprecated FAQ rich results in May 2026 and states structured data "doesn't affect how the page ranks." Microsoft's Bing team has confirmed schema helps Copilot parse content, but no platform publishes its citation logic, making any "proven formula" for ranking in ChatGPT or Perplexity an unverifiable guess. "AI visibility scores" vary wildly across tools because each uses its own prompt set and definition of a citation, with no industry standard.

The bottom line: ignore flashy percentage claims, treat any single visibility score as directional only, and focus on core content quality and retrievability — not schema hacks — as the only documented path to AI citation.

Open ten articles about "answer engine optimization" and you'll find the same move repeated: a suspiciously precise percentage, no methodology, no link, no author name attached to the underlying research. "FAQ schema increases AI citations by 40-60%." "Adding HowTo markup boosts ChatGPT visibility by 3x." These numbers get copied from post to post until they read as established fact, even though nobody can point to where they originated.

This matters more in AEO than in most corners of marketing, because the actual answer engines — Google's AI Overviews, ChatGPT, Perplexity, Copilot — don't publish their ranking or citation logic. When a claim about "what gets you cited" can't be traced to a disclosed algorithm or a transparent study, it isn't a finding. It's a guess wearing a lab coat.

This piece walks through the claims that circulate most, what's actually been tested or confirmed, and how to tell the difference next time you're evaluating an AEO claim — from us or anyone else.

Myth vs. fact

Myth (commonly repeated, rarely sourced)What's actually documented
"FAQPage schema makes content 40-60% more likely to be cited by AI"No published study supports a specific lift range. Google itself retired FAQ rich results for nearly all sites in 2023 and fully deprecated them in May 2026.
"Adding schema markup gets your page cited by AI"Ahrefs' matched-cohort study of 1,885 pages that added schema found no meaningful citation lift on Google AI Overviews, AI Mode, or ChatGPT after adding JSON-LD.
"Structured data is a ranking factor for AI Overviews"Google's own documentation states structured data "doesn't affect how the page ranks" — it only affects eligibility for certain rich-result formats, per Google Search Central.
"Our AI visibility score tells you exactly how you rank against competitors"There's no industry-standard methodology for these scores. Different tools query different prompt sets and weight citations differently, so scores aren't comparable across platforms.
"Schema markup does nothing for AI search"Overcorrection in the other direction. Microsoft's Bing team has publicly confirmed that schema helps its models (which power Copilot) parse content, and structured data still supports crawling, entity disambiguation, and traditional rich-result eligibility.
"There's a secret algorithm for ranking in ChatGPT/Perplexity we've reverse-engineered"Neither OpenAI nor Perplexity has published their retrieval or citation-ranking logic. What's public is directional (freshness, clarity, retrievability), not a formula — see the discussion of RAG-based citation mechanics from independent researchers below.

Why so many AEO claims can't be verified

Direct answer: The structural reason unverifiable stats dominate this space is simple: none of the platforms being optimized for disclose how they decide what to cite.

Google has stated, repeatedly and specifically for AI features, that there's no separate "AI index" or special AI-only markup — the same content that ranks in organic search is what AI Overviews draw from, evaluated by the same quality signals Google describes in its structured data guidelines. But Google does not publish the weighting between content quality, freshness, structured data, and the dozens of other signals that determine whether a specific passage gets pulled into a specific AI answer for a specific query.

ChatGPT and Perplexity are further behind on transparency. Neither company has published its retrieval or reranking pipeline. Industry researchers have built plausible external models of how Perplexity's citation pipeline likely works — query parsing, retrieval, multi-stage reranking, synthesis — but these remain inferred from observed behavior, not confirmed by the companies. Anyone selling you a "proven formula" for ranking in these systems is selling you their best guess, however well-informed.

This opacity is also why "AI visibility" scoring tools produce different numbers for the same brand on the same day. Each vendor picks its own prompt set, sampling frequency, and definition of a "citation" versus a "mention," and none of that is standardized industry-wide. Digiday has documented growing marketer skepticism about exactly this — inconsistent results across tools, no shared benchmark, and difficulty tying any of it back to revenue. That's not a reason to ignore visibility tracking; it's a reason to treat any single tool's score as directional within itself, not as an absolute number you compare against a competitor measured by a different tool.

What's actually documented

Direct answer: None of this means AEO is unknowable — it means the reliable claims are narrower and less flashy than the marketing copy suggests. Here's what holds up:

Structured data is not a ranking factor, but it does support machine understanding. Google is explicit that structured data doesn't move rankings directly; it affects eligibility for specific result formats and helps Google's systems parse entities and relationships on a page, per its own developer documentation. Separately, Bing's Fabrice Canel confirmed on record at SMX Munich in 2025 that schema markup helps Microsoft's LLMs (and therefore Copilot) understand page content — a real, attributable statement from an actual platform engineer, not an inferred claim.

Adding schema to already-well-cited pages doesn't produce a measurable citation bump. Ahrefs' difference-in-differences study is the most rigorous public test of this specific claim to date: pages that already had 100+ AI Overview citations before adding JSON-LD showed no statistically meaningful increase in citations afterward across Google AI Overviews, AI Mode, or ChatGPT — and a small statistically significant decrease on AI Overviews specifically. The authors are careful to note this doesn't rule out schema helping pages get crawled or indexed in the first place; it only tests the "add schema, get cited more" claim for content already inside the consideration set. Read the full study.

Google has been shrinking, not expanding, where FAQ schema visibly pays off. FAQ rich results were restricted to authoritative government and health sites in August 2023 and fully retired from Google Search in May 2026. Sites can keep the markup — Google has said unused structured data doesn't hurt you — but the visible payoff (the rich snippet itself) is gone for nearly everyone. Google's FAQPage documentation reflects this directly.

"How AI picks citations" is understood directionally, not mechanically. Across every credible source we found, the same non-secret advice repeats: be genuinely retrievable (crawlable, fast, not gated), answer the actual question clearly and early in the content, keep facts current, and earn real third-party references. This is closer to good editorial practice than a hackable formula, which is precisely why it's less exciting to sell as a "system."

How to evaluate the next AEO claim you read

Direct answer: A few checks take less than a minute and filter out most of the noise:

  • Is there a link to primary research, or just a number? If a stat has no attached methodology — sample size, date range, control group — treat it as marketing copy, not evidence.
  • Does the precision match the transparency of the underlying system? A claim like "40-60% more likely to be extracted" implies someone measured extraction probability across a real sample. Since no AI platform publishes citation logic, ask who could have run that measurement and how.
  • Is the vendor making the claim also the one selling the fix? Not disqualifying on its own, but it raises the bar for wanting to see the underlying data.
  • Does the claim survive contact with what the platforms themselves say? Google's own documentation and public statements are free to read and often directly contradict aggressive AEO marketing claims.
  • Is the number "portable"? An AI visibility score, share-of-voice percentage, or citation-rate figure that can't be reproduced with a stated prompt set, sample size, and date range isn't comparable to anyone else's number — including your own from a different tool.

None of this means AEO work is pointless — being clear, current, structured, and genuinely citable is good practice regardless of whether any specific tactic moves a specific AI platform's output. It means the claims should be sized to what's actually known, not to what's easiest to put in a headline.

Sources: