Last Tuesday I ran a GEO audit across 120 enterprise pages for a client in the fintech space. Brand mention was solid. Traditional rankings were top 3 for 78% of target terms. But when I checked those same terms in ChatGPT, Perplexity, and Google's AI Overviews? Only 16 pages got cited. That's 87% invisible in generative results.
The gap between ranking and being cited isn't closing. It's widening. And the 2025 audit playbook I used three months ago? Half of it doesn't work anymore.
Here's what I found, what I fixed, and what I'd stop doing immediately.
The citation gap got worse, not better
We published a breakdown of the citation gap earlier this year. The core problem: Google ranking ≠ AI citation. That's still true. But the gap is now structural, not just content-level.
I pulled citation data for 47 brand + topic queries across ChatGPT, Perplexity, and Gemini. Average citation rate for top-10 Google pages: 19%. Same pages from January 2024: 28%. That's a 32% drop in citation likelihood despite most pages getting *more* traditional traffic.
AI engines are getting pickier. They're not just scraping the SERP and citing whoever's on top. They're cross-referencing, they're weighting freshness harder, and they're pulling from sources that weren't even in the top 10.
One of my client's pages ranked #1 for "automated expense reporting for small business." Had been for 14 months. ChatGPT cited a Reddit thread from 2023 and a G2 review page. Not the #1 result. Not even close.
The page was . 3,200 words. Proper schema. Fast load time. Failed the AI visibility test completely.
What I tested: 6 GEO audit signals that actually moved the needle
I isolated six factors and ran them against citation presence. Ran the audit in the last two weeks of January 2025 across three AI engines. Sample: 120 pages, 47 queries, 5,640 individual citation checks.
Signal 1: Structured data coverage beyond Article schemaPages with FAQPage + HowTo + Product schema combined were cited 2.4x more often than pages with Article schema alone. This isn't new. But in 2025. the *combination* matters more than any single type. Pages missing FAQ schema specifically had a 12% citation rate. Pages with it: 31%.
Signal 2: Named entity density in first 400 wordsI measured how often the primary entity — brand name, product name, key person — appeared in the first 400 words. Citation rate jumped significantly when the entity appeared 3+ times in that window. Below 2 mentions? Citation rate dropped to single digits.
AI engines anchor on early entity signals. If your intro is vague or buries the brand mention, you're not getting cited.
Signal 3: Recency markers that aren't just dates in the URLThis one surprised me. I tested pages with "Updated [date]" in the body text vs. pages with just a schema dateModified vs. pages with neither. The body text update callout performed best. Schema-only date updates didn't move the needle at all.
AI engines aren't reading your structured data for freshness signals the way Googlebot does. They're reading the visible text. "Last updated January 2025" in plain text outperformed schema-only updates by a meaningful margin.
Signal 4: Third-party mention velocityI tracked how often each target page's key claims were mentioned on external domains in the 90 days before the audit. Pages whose claims were referenced on 5+ external domains in that window were cited at nearly double the rate of pages with low external mention velocity.
This is the hardest signal to manufacture. It's also the one that separates cited pages from invisible ones more than anything else.
Signal 5: Content depth on subtopics AI engines queryI pulled the actual subtopics AI engines ask about for each primary query. Then I checked whether the page covered those subtopics explicitly with H3-level sections. Pages covering 70%+ of AI-query subtopics were cited at a much higher rate. Pages covering under 40%? Single-digit citation rates.
This isn't "write more words." It's "answer the specific follow-up questions AI engines surface."
Signal 6: Citation format within the content itselfPages that included inline citations — linked references to data sources. studies, or official documentation — were cited 1.8x more often. AI engines prefer content that already behaves like a cited source. If your page doesn't reference external sources, AI engines treat it as lower-trust.
The fix that worked: a 3-week GEO retrofit
I picked 20 of the 120 pages — the ones with decent traditional traffic but zero AI citations — and ran a targeted retrofit.
Week 1: Entity optimization. Rewrote intros to front-load brand and product entity mentions. Added FAQPage schema to every page that lacked it. Inserted visible "updated" callouts.
Week 2: Subtopic coverage. Pulled AI-query subtopics for each page's primary term. Added 2-4 H3 sections per page covering the gaps. Each section 80-150 words — specific. direct, no fluff.
Week 3: External mention push. Sent product to three industry newsletter writers. Got two of them to include a mention with a link. Published one original stat on the blog and pitched it to four data-roundup curators.
Results after 6 weeks: citation rate across the 20 pages went from 0% to 34%. Three pages started appearing in AI Overviews for their primary terms. One page — a guide on expense policy templates — started getting cited in ChatGPT for four related queries it didn't even target.
Total time investment: about 11 hours. Not nothing. But not a site rebuild either.
What I'd stop doing for GEO in 2025
Stop optimizing for "AI-friendly content" as a vague concept. There's no such thing as generically AI-friendly content. There's content that answers specific subtopics AI engines query, with proper entity signals, freshness markers, and external validation. That's it. Stop assuming schema alone will get you cited. Schema helps Google understand your page. AI engines are reading the rendered text. Your visible content matters 10x more than your structured data for citation purposes. I know that's heresy to some SEOs. The data doesn't care. Stop ignoring AI Overview cannibalization. If your page is ranking #1 but getting zero clicks because an AI Overview answers the query above it. you're not winning. You're losing slowly. Check your zero-click search metrics. If click-through on position 1 is below 50% for a query, an AI Overview is probably eating your traffic. Stop treating GEO as separate from technical SEO. Page speed still matters. AI engines crawl and render. If your LCP is over 4 seconds. you're losing crawl budget and rendering quality. Fix your Core Web Vitals. They're not just for Google rankings anymore.The uncomfortable reality of GEO audits in 2025
Most GEO audit reports I see from agencies are useless. They're keyword-difficulty analysis repackaged with "AI" in the title. They check whether you have schema. They check whether your content is "." None of that predicts citation presence.
What predicts citation presence is: entity density in visible text, subtopic coverage matching AI-query patterns, external mention velocity, and inline source citations. Four signals. That's the audit. Everything else is noise.
The new SERP reality means your audit framework from 2023 is actively misleading you. Update it or stop calling it a GEO audit.
I'll keep running these tests monthly. The signals shift. What worked in November didn't work in January. That's the pace now.
Frequently Asked Questions
How often should I run a GEO audit?Every 6-8 weeks minimum. AI engine behavior shifts faster than traditional algorithm updates. If you audited in October 2024. your findings are already stale.
Does getting cited in AI search actually drive traffic?It depends on the engine. ChatGPT citations drive direct traffic. Google AI Overviews citations mostly drive brand searches. Perplexity sits in between. Measure branded search volume changes, not just referral traffic.
Is GEO replacing traditional SEO?No. They're converging. The pages that win in AI citations are usually strong traditional SEO pages with specific GEO retrofits layered on top. You can't skip traditional optimization and expect AI visibility.
Which AI engine should I optimize for first?Whichever one your actual users are querying. B2B audiences lean Perplexity and ChatGPT. Consumer queries lean Google AI Overviews. Check your analytics for AI referral sources before picking a target.
References
Take this with a grain of salt — this is just my experience. If you disagree. you are probably right.