Expert quotes and statistics are the two content elements most strongly associated with AI citation: pages that contain an attributed quote from a named, credentialed expert are cited noticeably more often than equivalent pages without one, and pages that contain a specific, sourced statistic see a similar lift. Both work for the same underlying reason - they give a large language model a discrete, verifiable claim it can extract and attribute, rather than a vague generalisation it has to interpret and paraphrase on its own.
This matters because most business content is written the opposite way. It is full of adjectives, hedged claims, and unattributed assertions - "many experts agree", "studies show", "this approach works well" - that read fine to a human skimming for tone but are useless to a model trying to extract a fact it can stand behind. If you want to understand why some pages get pulled into ChatGPT, Perplexity, and Gemini answers while near-identical competitors do not, the presence and quality of quotes and statistics is one of the clearest differentiators in the data.
Why AI models are extractive readers, not interpretive ones
When an AI engine composes an answer, it is not summarising your overall argument the way a human editor would. It is retrieving passages, scoring them for relevance, and deciding which ones to lift into its response, usually with an attribution back to the source. A passage built around a vague claim forces the model to either omit it, soften it further, or take on liability for a claim it cannot verify. A passage built around a named expert's statement or a specific number does none of that - the model can quote or paraphrase it and point to exactly who or what is responsible for the claim.
This is also why generic marketing copy performs so poorly in AI citation, even when it targets the right keywords. "Our platform delivers industry-leading performance" contains no extractable fact. "Our platform processes 2.3 million transactions per second in independent benchmark testing conducted by [named lab] in March 2026" contains a number, a source, and a date - three things a model can check, attribute, and safely reuse.
The anatomy of a citable quote
Not every quote helps. A quote from an anonymous "industry expert" or an unnamed "senior analyst" is barely more useful to a model than no quote at all, because the model cannot establish who is making the claim or why they are qualified to make it. A citable quote needs four things working together.
- A named, real person - full name, not a title alone
- A stated credential or role that establishes why this person is qualified to make the claim
- A specific, opinionated, or factual statement - not a generic platitude that could apply to any topic
- Correct attribution formatting, ideally with Quotation or Person schema markup, so the claim and its source are machine-readable, not just visually adjacent
The credential requirement extends beyond a single quote. If the person being quoted is also the named, consistent byline author of other content on your site, that name compounds in weight every time it reappears with matching schema and a linked bio page. See our breakdown of whether named authors get cited more by AI for the data on how byline attribution functions as its own citation signal, separate from any individual quote.
The anatomy of a citable statistic
The same logic applies to numbers. A statistic without a source, a date, or a defined methodology is a claim a model has to treat with suspicion, because it cannot verify where the number came from or whether it is still accurate. A statistic with all three is a fact the model can lift with confidence.
- A precise figure, not a rounded generality - "37% of respondents" rather than "over a third"
- A named source - the study, survey, or dataset the number came from
- A date, so the model and the reader can judge how current the figure is
- A defined sample or methodology where relevant - who was surveyed, how large the dataset was, what was measured
If you cannot name the source of a statistic you are about to publish, do not publish it as a bare number. Either find the primary source and cite it properly, or rephrase the claim without the number. An uncited statistic is a liability for both AEO and basic editorial credibility.
How to source real quotes without inventing them
The temptation when a page needs a quote is to write one and attribute it loosely, or worse, to fabricate a plausible-sounding expert. Both are editorial and, increasingly, legal risks, and fabricated attribution is also fragile from an AEO standpoint: a model or a fact-checking layer that cannot verify a named person or their stated role is less likely to treat the claim as trustworthy, and more likely to skip the page entirely.
- Interview your own team - founders, engineers, and customer-facing staff usually have specific, defensible opinions that make good quotes once written down properly
- Reach out to customers or partners for a short, specific comment rather than a generic testimonial
- Use a journalist-request platform to source quotes from external experts on a specific claim
- Quote from a published interview, talk, or paper, with a link back to the original source
- Record and lightly edit a real conversation rather than paraphrasing from memory - specificity survives better when transcribed
Where to place quotes and statistics for maximum extraction
Placement interacts directly with how AI models chunk a page. A brilliant statistic buried in paragraph fourteen of a long article may never be retrieved, because retrieval systems weight the earlier portion of a page more heavily and because a chunk deep in the page competes with everything above it for the model's attention.
Lead with the strongest data point
Your single best statistic or quote should appear in the first two or three paragraphs, ideally in the same passage as your definition-first opening sentence. This gives the model an extractable, attributable fact the moment it starts scoring your content, rather than several hundred words into an argument it may never fully retrieve.
Repeat supporting evidence inside FAQ and section answers
Individual FAQ entries and section subheadings are each treated as near-independent retrieval units. A quote or statistic that only appears once, in the introduction, will not surface if the model retrieves a later section instead. Distribute your strongest evidence across the sections and FAQs most likely to be retrieved on their own, not just at the top of the page.
Do not recycle the same three-year-old statistic across dozens of pages. AI models increasingly cross-reference dates, and a widely repeated but stale figure erodes trust in every page that uses it. Refresh or replace ageing statistics as part of your normal content maintenance cycle.
Common mistakes that make quotes and statistics uncitable
- Attributing a quote to a title only ("a company spokesperson said") instead of a named person
- Publishing a statistic with no link to its source, date, or methodology
- Burying strong evidence in the final third of a long article where it is rarely retrieved
- Using the same quote or statistic on every page of the site, which reduces distinctiveness and can read as low-effort duplication
- Rounding numbers so heavily that the figure loses its specificity - "most users" instead of "78% of surveyed users"
- Letting a cited statistic go stale for years without a refresh or a removal
A before and after example
Before: "Many businesses have found that answer engine optimisation improves their visibility in AI search results, according to industry experts." This sentence contains no named source, no number, and no verifiable claim. A model cannot extract anything from it beyond a vague sentiment.
After: "'Sites that add FAQPage schema to their top ten pages typically see a citation increase within six to eight weeks,' says Maria Chen, Head of SEO at a mid-market SaaS company that ran the change across 40 client sites in Q1 2026." This version names a person, states a role, and gives a specific, time-bound claim tied to a described sample. It is longer, but every additional word adds extractable, attributable information rather than filler.
Measuring whether it is working
Track AI referral traffic in GA4 before and after you add attributed quotes and sourced statistics to a page, and periodically ask ChatGPT, Perplexity, Claude, and Gemini the exact questions the page answers. Watch for the specific quote or figure appearing in the model's response, even when your URL is not the cited source - that is a sign the underlying claim is propagating through AI retrieval systems, which is the goal even before direct citation catches up.