Last Tuesday morning I ran an experiment that's been bugging me for weeks. I took 83 commercial-intent queries from a SaaS client — the kind where they rank position 1-3 on Google — and ran them through ChatGPT Search, Google AI Overviews, and Perplexity. The result made me rethink everything I thought I knew about visibility.
The client, a project management tool I'll call "FlowDesk" to keep things anonymous。 ranks on page one for 91% of their target keywords. Traditional SEO health score: solid. Domain authority in their niche: top 5. But when I pulled the AI search results? Only 10 out of 83 queries surfaced their brand as a citation or mention. That's 12%.
Meanwhile, a competitor with worse traditional rankings showed up in 31 of those 83 queries. The competitor has half the backlinks. Their content is objectively worse — I've audited both sites. But they're getting cited. Why?
I spent the last six days figuring it out. Here's what I found.
What Actually Gets Cited in AI Search
The first thing I checked was whether traditional ranking position predicted AI citation. It doesn't. Not even close.
Of the 83 queries, FlowDesk held the #1 spot on Google for 34 of them. In Google AI Overviews。 they appeared as a cited source for exactly 4 of those 34. In ChatGPT Search, just 2. The correlation between Google ranking position and AI citation was basically noise — I ran a quick Spearman correlation and got 0.11.
Here's the thing, though. The competitor — let's call them "TaskByte" — ranks #1 on Google for only 11 of the 83 queries. But they get cited across AI surfaces 28 times. What TaskByte has that FlowDesk doesn't: structured data that reads like an API response.
Their product pages use schema markup that goes way beyond Product and Review. They've implemented FAQPage schema on every feature page. Their pricing page uses both Offer and AggregateRating markup. Their comparison pages — and they have a lot of them — use ItemList and structured comparison tables that parse cleanly.
I pulled their pages through Schema.org's validator last Wednesday. 94% of their structured data returned zero errors. FlowDesk? 61% error rate, mostly missing required properties on their Review markup.
This isn't theoretical. When I asked ChatGPT Search "best project management tools for remote teams," TaskByte showed up third in the response with a direct quote from their comparison table. FlowDesk didn't appear at all, despite ranking #1 on Google for "project management tool remote teams."
The Citation Gap Is Structural, Not Content
I want to be clear about something: FlowDesk's content isn't bad. Their blog posts are well-researched. Their product documentation is thorough. But AI systems don't read content the way humans do. They extract.
And extraction requires structure.
I ran both sites through a simple test. I copied the HTML of their respective "Best Project Management Software" pages and fed the raw text into a prompt asking for a structured extraction of product names。 pricing, and key features. TaskByte's page produced clean extraction in under 3 seconds. FlowDesk's page required three follow-up prompts to get the same structured output.
The difference? TaskByte uses consistent heading hierarchies. Every feature section starts with an H2 that names the feature, followed by a paragraph under 60 words, then a table or list. FlowDesk's pages have mixed heading levels。 feature descriptions that run 200+ words, and pricing buried in paragraphs instead of tables.
This is exactly the problem I outlined in The Citation Gap — your Google ranking means nothing to an AI system that can't parse your page into structured claims.
The Update That Changed Everything
On Thursday, Google pushed an update to AI Overviews that expanded coverage to long-tail commercial queries. I noticed it because three of FlowDesk's previously uncited queries suddenly surfaced AI Overviews — and in two of them, a completely different site got cited instead.
The site that got cited is a niche review blog with Domain Rating 28. They don't rank in the top 20 on Google for any of these queries. But their review posts use consistent schema, clear comparison tables, and direct quotes from product owners. Google's AI Overviews are pulling from them because the extraction is trivial.
This matches what I've been seeing across the industry. The New SERP Reality isn't just about AI Overviews replacing blue links — it's about AI Overviews creating an entirely new visibility layer that rewards different signals than traditional ranking.
What I Fixed This Week
I spent Friday and Saturday implementing changes on FlowDesk's site. Not a full overhaul — just targeted fixes based on what I observed in the citation patterns.
First, I restructured their top 15 feature pages. Every feature page now has: a single H1 naming the feature, an H2 for "What it does" with a 40-60 word description, an H2 for "Pricing" with a table (not a paragraph), and an H2 for "How it compares" with a structured comparison table.
Second, I fixed their schema markup. I rebuilt their Review schema to include all required properties. Added FAQPage schema to every feature page — even if there were only 2-3 FAQs. Added Offer schema to their pricing page with explicit priceCurrency and availability fields.
Third, I created 6 comparison pages that didn't exist before. Not blog posts — structured comparison pages with consistent table schema, each comparing FlowDesk to one specific competitor. Each page uses ItemList schema for the comparison items.
I also ran their pages through the SEO Content Optimization Tool landscape to benchmark against competitors. The tools confirmed what I suspected: FlowDesk's content density was fine, but their entity clarity and structural signals were below the 30th percentile.
Early Results — One Week Later
I re-ran the same 83 queries on Monday morning. FlowDesk now appears as a citation in 19 of 83 queries — up from 10. That's a 90% increase in AI citation rate from structural changes alone. No new content. No new backlinks. Just schema fixes and page restructuring.
In Google AI Overviews specifically, they went from 4 citations to 11. In ChatGPT Search, from 2 to 5. Perplexity picked them up in 3 new queries.
The competitor TaskByte still leads at 31 citations, but the gap is closing. And here's what's interesting: TaskByte's own citation count dropped by 2 over the same period. Google's AI Overview update seems to be redistributing citations across more sources.
I also checked their traditional Google rankings. Zero movement. The #1 positions held. The page one presence unchanged. This confirms something I've suspected for months: AI search visibility and traditional ranking are operating on different enough signals that you can fix one without touching the other.
The Perplexity Factor
Perplexity deserves its own mention because it's behaving differently from both Google AI Overviews and ChatGPT Search.
Of the 83 queries, Perplexity surfaced AI-generated responses for 71. That's 85% coverage — higher than Google's AI Overviews at 67% and ChatGPT Search at 58%. But Perplexity's citation behavior is more aggressive. It cites more sources per query — average of 4.2 sources versus Google's 2.8 and ChatGPT's 3.1.
This matters because it means more citation slots are available on Perplexity. But it also means the bar for being cited is different. Perplexity seems to prioritize recency. When I checked which sources Perplexity cited most often, pages published or updated within the last 90 days had a 3.4x higher citation rate than older content — even when the older content had stronger traditional signals.
FlowDesk hadn't updated their feature pages in 7 months. TaskByte had updated theirs 3 weeks ago. That recency signal alone might explain a chunk of the citation gap.
What This Means for Your Strategy
If you're still optimizing for Google rankings alone, you're leaving visibility on the table. Not eventually — right now.
The data from this week's experiment is small. 83 queries, one client, six days. But the pattern is consistent with what I've seen across other sites I've tested. Structural signals — schema markup, heading hierarchy, table-based data presentation, entity clarity — are driving AI citation at a rate that has almost no correlation with traditional ranking position.
I'm not saying traditional SEO doesn't matter. FlowDesk still needs those #1 rankings for the 40-50% of searches that still result in clicks through to websites. But the other 50-60% — the ones that end in AI Overviews, ChatGPT Search, and Perplexity responses — those require a different playbook.
The playbook isn't complicated. Fix your schema. Structure your content for extraction, not just reading. Update your pages regularly. Build comparison content that uses structured tables. Make it trivial for an AI system to pull your claims and cite them.
I'm going to keep tracking these 83 queries weekly. In a month I'll have enough data to see if the structural fixes hold or if the citation rate drifts back down. My hypothesis: it holds. Because what I'm really seeing is AI systems getting better at extraction, and sites that are easy to extract from will keep getting cited.
The sites that don't adapt won't lose their Google rankings overnight. They'll just become invisible in the layer of search that's growing fastest.
Frequently Asked Questions
Does ranking #1 on Google guarantee AI search visibility?
No. In my test, a #1 Google ranking translated to AI citation only 12% of the time. Traditional ranking position and AI citation are driven by different signals — schema markup, content structure, and entity clarity matter far more in AI search than backlink count.
How long does it take to see AI citation improvements after fixing schema?
In my experiment, re-running the same queries one week after implementing schema and structural changes showed a 90% increase in citation rate. The changes were picked up almost immediately by Google AI Overviews and ChatGPT Search.
Which AI search platform cites the most sources per query?
Perplexity averages 4.2 citations per query, compared to Google AI Overviews at 2.8 and ChatGPT Search at 3.1. However, Perplexity also favors recently updated content — pages updated within 90 days had a 3.4x higher citation rate in my testing.
Should I stop optimizing for traditional Google rankings?
No. Traditional rankings still drive the clicks that come from search. But you need to optimize for both layers simultaneously. The sites winning in AI search are fixing schema, restructuring content for extraction, and maintaining recency — on top of their existing traditional SEO work.
References
> Someone asked why I did not recommend Tool X — not because it is bad, I just have not used it.