Does adding statistics and quotes make AI cite my content?

Adding statistics and quotes as a citation tactic has weak evidence, and the best-designed tests point the other way. The advice comes from lab work where the page was already in front of the model. In live AI search, your page first has to be fetched, indexed and retrieved for the query. Only then does wording matter.

It helps to split the process into four stages:

• Parametric: what the model knows with no web access. Nothing you publish this quarter reaches it.
• Access: whether a crawler can fetch and store your page. You control most of this.
• Retrieval: whether your page enters the model's context for a given query. The engine controls this.
• Generation: which retrieved pages get used and cited. The model controls this.

Statistics and quotes are generation-stage tactics. A critical survey of 45 GEO studies (arXiv, July 2026) concluded that already-retrieved content "can causally alter its citation or use, but no reviewed technique shows a stable, longitudinal, cross-platform causal effect on organic discoverability." So the question for most brands isn't phrasing. It's whether the engine ever sees the page.

The 2023 GEO paper (Aggarwal et al., published at KDD 2024) is the usual source for "add statistics and quotations." Its setup matters more than its headline.

• The engine was the authors' own. It was a two-step system with the sources supplied in context. No live index was queried.
• The "in the wild" test wasn't quite wild. For Perplexity, the authors uploaded source text as files, because perplexity.ai does not let users specify source URLs.
• The metric is circular. Position-Adjusted Word Count counts words from a source in the answer, including words the intervention itself added.
• The famous numbers are not in the paper. It reports a 30 to 40% band across three methods, not the itemised figures that circulate in GEO marketing.

The paper also found keyword stuffing decreased its own visibility metric (17.8 against a 19.5 baseline). That is a useful warning: tactics that look like optimization can backfire even in a friendly lab.

For how these setups differ from production engines, see lab GEO studies vs live AI engines.

C-SEO Bench (NeurIPS 2025) tested conversational SEO methods across several models. Its authors reported that "most current C-SEO methods are not only largely ineffective but also frequently have a negative impact on document ranking." Moving a document to the top of the model's context was far more effective. That's a positional effect no publisher can invoke. The same study found gains shrank as more sites adopted the tactics, which it described as zero-sum.

A 252,000-trial factorial study (arXiv, 2026) found "formatting-only edits have little impact." It named topical relevance and list position as the biggest drivers of being cited first. It also found that "including explicit price information and a recent timestamp also helps consistently." Read that with care: it was industry-authored, used a synthetic e-commerce corpus, and gave the model only two candidate documents per trial.

The July 2026 survey adds that "citation-oriented rewrites can impair retrieval." None of this proves content quality is irrelevant. It shows that rewriting a page to look quotable is not a reliable route into answers. The comparison of on-page rewrites and access fixes sets the two evidence bases side by side.

Two things get lumped together under "add statistics." They behave very differently.

Stating real, specific facts

A page that says your starter plan costs a stated amount per month, or that your clinic opened in a given year, gives an engine a sentence it can quote. That is the only positive experimental signal for on-page content in the evidence. It's not a guarantee. It just means that when your page is retrieved, the fact is there in plain words.

Say you run a hypothetical accounting firm. "We file returns for 400 small businesses a year" is a fact if it's true and current. Pasting an industry statistic with a vague attribution into every page is decoration.

Injecting statistics or quotes as a trick

The evidence base treats appended quotation blocks, injected statistics and "according to authoritative sources" boilerplate as risk flags. The reasons stack up:

• Fabricated or stale numbers can get repeated by AI as facts about your brand.
• Google says you don't need to write in a specific way for generative AI search.
• Mass-produced filler can drift into territory covered by Google's policies. See whether GEO tactics can violate Google's spam policies.
• Hidden text or prompt injection aimed at AI crawlers is illegitimate. It is manipulation, and it carries reputational risk if exposed.

If earlier agencies added this kind of material, the playbook for auditing risky GEO tactics walks through finding it.

Work on the parts you control, in this order.

1. Make the facts reachable. Vercel's crawler study found ChatGPT and Claude crawlers fetch JavaScript files but don't execute them. If your prices or product names appear only after JavaScript runs, many AI crawlers never see them. Put them in the server-rendered HTML.
2. Check per-bot access. Robots.txt rules are per bot. OAI-SearchBot controls ChatGPT search, while GPTBot governs training only. A firewall or CDN can also block a crawler that robots.txt allows.
3. State each key fact once, clearly. One plain sentence per fact beats a block of borrowed statistics. The guide on what makes a sentence easy for AI to quote covers this in more detail.
4. Keep facts consistent. The same price written differently on your pricing page and footer gives engines two answers to pick from.
5. Look at what engines say now. The free checker on this site shows whether AI names you for up to three buyer questions. It's a snapshot, not tracking.

None of these steps promises a citation. They remove the blocks that stop engines from using your content at all.

Should I remove statistics and quotes already on my pages? Not if they're real, sourced and useful to readers. The evidence doesn't show that genuine data hurts. It shows that adding statistics or quotation blocks purely to game citation doesn't reliably work and can backfire. Remove fabricated, unsourced or stale numbers, and boilerplate added only for AI. Keep the facts a customer would actually want.

Where does the claim that GEO adds up to 40% visibility come from? It traces to the 2023 GEO paper, which reported a 30 to 40% band across three methods. That band was measured on an engine the authors built, with sources already in context, using a metric that counts words the intervention itself added. It isn't evidence about live AI search, and it doesn't describe what happens to a real site.

Does citing my sources help AI cite me? No study in our evidence base isolates outbound citations as a lever for live AI answers. Linking to sources is good practice for readers and for your credibility. Treat it as part of writing well, not as a mechanism that gets your page retrieved or quoted by an engine.

Do these findings apply to the newest AI models? Unclear. As of October 2026, the main rewriting studies, including C-SEO Bench, ran on 2025-era models and have not been replicated on the current frontier releases. The direction of the evidence is consistent across studies, but nobody has published a fresh test. Plan around access and accurate facts, which don't depend on a model version.