Two companies in the same niche. Similar organic rankings. One gets named by ChatGPT and Perplexity constantly; the other is invisible in AI answers. Same keywords, opposite outcomes. In the audits I run, this is the pattern I keep seeing—and the cause is almost never ranking position.
The pattern that started this
When a client asks why a competitor keeps showing up in AI answers, my first instinct used to be “they must rank higher.” They usually don’t. What actually separates the two, once you read the pages side by side, is editorial: one page quotes real people, cites real numbers, and links out to primary sources. The other makes claims and hopes you believe them. Generative engines treat those two pages very differently.
There’s now peer-reviewed research that measures exactly how differently. And the good news for a B2B team with limited editorial capacity: the highest-impact changes are things you can do to an existing article in an afternoon.
GEO isn’t SEO: ranking is not a citation signal
Search engine optimization gets you a position in a list of ten blue links. Generative engine optimization (GEO) gets you named inside a synthesized answer. Those are different games with different winners, and the clearest evidence is a March 2026 Ahrefs analysis of 863,000 keyword SERPs and roughly 4 million AI Overview URLs.
Ahrefs found that only 37.9% of URLs cited in Google’s AI Overviews also appeared in the top 10 organic results. Flip that around: about 62% of cited pages don’t rank in the top 10 at all. An earlier 2025 pass had put the overlap near 76%, so the ground is shifting—but the direction is unambiguous. Being on page one is neither necessary nor sufficient to get cited.
Where the percentages come from
Almost every credible GEO number traces back to one source: a paper titled GEO: Generative Engine Optimization, presented at KDD ‘24 (the ACM SIGKDD conference, Barcelona, August 2024) by researchers from Princeton and IIT Delhi. It introduced GEO as “the first novel paradigm to aid content creators in improving their content visibility in generative engine responses,” and it built a benchmark—GEO-bench—of 10,000 queries spanning nine source types and 25 domains.
That scale matters, because it’s what lets the authors say something concrete: the top methods “achieved a relative improvement of 30–40% on the Position-Adjusted Word Count metric.” Position-Adjusted Word Count (PAWC) is their proxy for visibility—it rewards content that appears, and appears early, in the generated answer.
Here’s the part most blog posts get wrong, so I’ll be precise. The study’s Table 1 reports PAWC scores, not percentage lifts. Adding direct quotations scored 27.8. Embedding statistics scored 25.9. Citing sources scored 24.9. Unoptimized content scored 19.5. Against that baseline, those are relative gains of roughly 43%, 33%, and 28%—which is exactly why the paper caps its headline at “up to 40%.” If you see “+27.8% lift” somewhere, someone confused a raw score for a percentage. The three tactics are real and they’re the strongest in the study; the framing just needs to be honest.
Two more things keep me confident in these numbers. The authors re-ran the methods on Perplexity.ai—a live, public generative engine, not just their lab setup—and quotation addition still delivered a 22% improvement on PAWC. And they showed a tactic can backfire: keyword stuffing scored 17.8, below the 19.5 baseline. So this isn’t “do more of everything.” It’s specific.
Tactic 1—embed direct quotes (the strongest single move)
Quotation addition topped the study. The logic is intuitive once you say it out loud: a quote is a verifiable, attributable statement, and generative engines lean on content that looks like it can be trusted and repeated.
For a B2B service business, three kinds of quotes carry the most weight:
- Expert voices—a named specialist on your team, or an outside authority you’re paraphrasing accurately with attribution.
- Customer language—a real client sentence about the problem you solved, not a scrubbed testimonial.
- Regulatory or standards bodies—the exact wording from a law, standard, or official guideline, quoted rather than summarized.
Structure them so a machine can lift them cleanly: a full sentence in quotation marks, the speaker named, and enough surrounding context that the quote stands on its own if it’s the only thing pulled. A practical example for a services blog: instead of “onboarding is often the bottleneck,” write—Our operations lead put it bluntly: “Every delayed launch we reviewed last year traced back to onboarding, not the build.” That’s a sentence an engine can quote and attribute. The vague version isn’t.
Tactic 2—statistics with a source link, not bare claims
Statistics addition scored 25.9 in the study, and on Perplexity it showed gains of up to 37% on the subjective-impression metric. But there’s a catch the research implies and my audits confirm: an unsourced number reads as a liability, not an asset. A figure with no link behind it is a claim a careful system has no reason to repeat.
Compare two versions of the same paragraph.
Before: “Most AI-cited pages don’t even rank on page one—ranking barely matters anymore.”
After: “A March 2026 Ahrefs study of 863,000 SERPs found only 37.9% of AI Overview citations came from the top 10 organic results—so most cited pages don’t rank on page one.”
The second version gives a number, a date, a sample size, and an attributable source. That’s the shape of a sentence that gets quoted. The first is an opinion. And remember keyword stuffing scored below baseline—padding a page with numbers you can’t back isn’t a substitute for a few you can.
Tactic 3—explicit source citations as a trust anchor
Citing sources scored 24.9, and it comes with the most interesting finding in the whole paper: the effect is asymmetric. For a page ranked fifth in Google, adding source citations produced a 115.1% increase in AI visibility. Meanwhile the top-ranked page saw its relative AI share drop by about 30.3%.
Read that again if you’re not in the organic top three, because it’s aimed at you. Outbound links to authoritative primary sources are one of the few levers that lets a mid-ranked B2B page punch above its SERP position in AI answers. The source types that carry the most weight: primary research and academic papers, official standards and regulatory texts, government data, and named-author industry studies. A link to a competitor’s marketing blog does little; a link to the underlying dataset does a lot.
This is also where measuring becomes non-optional. If ranking and AI citation have decoupled, your rank tracker can’t tell you whether any of this is working. We built Cited for exactly this gap—it tracks how ChatGPT, Claude, Perplexity, Gemini, and Google AI Overviews describe your brand and competitors over time, so you can see citation share move after an edit instead of guessing.
The freshness factor
Editorial quality gets you eligible to be cited. Freshness decides how long you stay in the pool. Research from Lily Ray and the team at Amsive, reported by SalesPeak, found that 50% of the content cited in AI answers is less than 13 weeks old—framed not as an algorithmic preference but as “a structural feature of how AI search works.” (That’s a secondary citation, so treat it as directional rather than gospel.)
Amsive’s own analysis shows why. New content can capture up to 2% of all citations in a niche within 2–3 days of publishing, then decays to about 0.5% within one to two months—roughly a 75% drop. The freshest pages dominate the citation pool, which is exactly why half of citations skew under 13 weeks old.
For a B2B team without a newsroom, the takeaway is a relief, not a burden: cadence beats volume. You don’t need to publish weekly. You need a substantive quarterly refresh of your most important pages—new stat, new quote, a genuinely updated section, and a real change to the modified date. That keeps your best assets inside the window where they get cited.
Topic targeting beats keyword density
Because keyword stuffing scored below baseline, the old density game actively hurts here. What helps is semantic depth—covering a topic thoroughly enough that your page can answer many phrasings of the same underlying question, not just one exact-match query.
In practice, that changes how you outline a technical article. Instead of one page per keyword, build one deep page per question a buyer actually asks, and structure it so each sub-question gets its own clearly headed, self-contained section. Generative engines assemble answers from passages; a page organized as a set of clean, quotable passages gives them more surfaces to pull from than a page optimized to repeat a phrase.
Implementing this in WordPress
None of this needs a developer. In the block editor:
- Use the Quote block for embedded quotations—it produces clean, semantically marked-up markup an engine can parse, and it visually signals “this is quotable.”
- Put statistics in their own short paragraph or a list item with the source linked inline, rather than burying the number mid-sentence.
- Add outbound source links as normal inline links to primary sources; you don’t need nofollow on genuine citations.
- Use a schema plugin (Yoast, Rank Math, or similar) to emit Article schema with a named author, so the E-E-A-T signal is machine-readable.
- Make sure your refresh actually updates dateModified—some setups only touch it on structural edits, so confirm the field changes when you update content.
If your hosting or theme fights you on any of this—stale modified dates, broken schema, slow rendering that keeps crawlers from seeing your content—that’s a foundation problem worth fixing before you chase citations.
The GEO content-refresh checklist
Ten things that turn an existing article into a citable one, in about an hour:
- Add at least one attributable direct quote—expert, customer, or regulatory—as a full quoted sentence with a named source.
- Find every bare claim with a number and either link it to a real source or soften it to something you can defend.
- Add one outbound link to a primary source (research, standard, official data), especially if the page isn’t in the organic top three.
- Remove keyword repetition that exists only for density—it scored below baseline in the research.
- Break long sections into clearly headed, self-contained passages an engine can lift.
- Add or tighten a short, direct answer near the top for the main question the page targets.
- Confirm the byline names a real author and that Article/Person schema is emitted.
- Update the publish or modified date only if you’ve made a substantive change—don’t fake freshness.
- Re-read for marketing filler and cut it; specific beats persuasive here.
- Note the date and set a reminder to revisit in one quarter.
These tactics are measurable—which is the point
Everything above is testable, and the payoff isn’t vanity. In a Seer Interactive case study of one B2B client (October 2024–April 2025), visitors arriving from ChatGPT converted at 15.9% versus 1.76% for Google organic—roughly 9×. It’s a single client, and AI traffic was still a tiny share of the total, so I won’t oversell it. But the direction lines up with what we see: AI-referred visitors arrive further along in their decision, and getting cited puts you in front of them.
That’s exactly what we check in the AI Visibility Audit—where you’re cited today, where competitors are winning, and which of these editorial changes will move the needle on your specific pages. If you’d rather talk it through first, get in touch. The tactics are proven and the effect is measurable. The only question is whether your best pages are doing any of it yet.