I tracked 200 pages through AI search results for 30 days — here's what actually moved the needle

Last week I pulled our content performance data and noticed something weird: our organic traffic from Google was flat, but our referral traffic from ChatGPT and Perplexity had jumped 340% month-over-month. Same content. Same pages. Completely different distribution channels.

I've been running a small experiment for the past 30 days tracking how 200 of our pages perform across generative AI search engines versus traditional SEO. The results are messy, counterintuitive, and honestly a bit frustrating if you're still optimizing only for blue links.

I picked 200 pages from our blog — a mix of how-to guides, opinion pieces, and product docs. I wanted to see which ones got cited in AI overviews, which ones got completely ignored, and whether traditional SEO signals (backlinks, domain authority) actually correlate with AI visibility.

I checked each page manually in ChatGPT, Perplexity, and Google's AI Overviews for 15 different queries. That's 200 pages × 3 engines × 15 queries = 9,000 checks. Tedious? Yes. Necessary? Also yes, because the automated tools I tried were missing nuance.

I tracked three things for each page:

1. Citation frequency: How often the page appeared in AI responses (0-15 scale)

2. Extraction quality: Whether the AI pulled the right information or mangled it

3. Traditional SEO metrics: DR, backlinks, ranking position

Pages with clear, scannable structure won. Not "good content" in the abstract — specifically pages with:

Our top-performing page in AI results was a 1,200-word guide on API rate limiting. It ranks #8 on Google for the target keyword but shows up in 73% of AI responses for related queries. Meanwhile, a 3,000-word "ultimate guide" that ranks #2 on Google? It gets cited in maybe 12% of AI responses.

The difference: the shorter page leads with a concrete example. The longer page buries the lead under three paragraphs of introduction.

Here's where it gets interesting. I ran a correlation analysis between traditional SEO metrics and AI citation frequency. The correlation between Domain Rating and AI citations was 0.31 — weak. The correlation between page structure clarity (measured by something I'm calling "answer density" — how quickly you get to the point) and AI citations was 0.67.

Backlinks still matter for Google. But for AI engines, they're reading your content differently. They're not counting votes; they're extracting answers.

I also found that pages with tables and structured data performed 28% better in AI extraction. If you're presenting comparative information in paragraph form, you're making it harder for AI to pull out the key points.

After identifying the pattern, I restructured 20 of our worst-performing pages. I didn't rewrite them — I just:

1. Moved the key takeaway to the first 50 words

2. Broke up paragraphs longer than 3 sentences

3. Added specific numbers where we had vague claims

4. Added clear attribution for any data cited

5. Converted long comparative sections into tables

Within two weeks, AI citations for those 20 pages increased by an average of 45%. Google rankings stayed roughly the same. The effort was maybe 30 minutes per page.

One specific example: Our "Email Deliverability Guide" was a 2,800-word monster that ranked #4 on Google but got zero AI citations. I restructured it to lead with a specific stat ("94% of emails reach inbox with proper authentication"), broke up the intro, and added a comparison table. AI citations went from 0 to 8 out of 15 queries. Google ranking stayed at #4.

I tried a few tools to automate this analysis. Most GEO tools I tested gave me a binary "optimized/not optimized" score without explaining why. The GEO Audit Tool was actually useful because it showed me exactly which sections of my content were being extracted by AI engines, not just a score.

For understanding whether your content has the "gravity" to pull in AI citations, the AI Gravity Checker gave me a more nuanced view than traditional SEO metrics. It's not perfect, but it's closer to what I actually care about now.

If you're still treating AI search as a side channel, you're making a mistake. The GEO vs SEO landscape is shifting faster than most people realize. I'm not saying abandon traditional SEO — I'm saying the optimization criteria are diverging.

Google wants comprehensive, authoritative, link-rich content. AI engines want concise, structured, answer-first content. These aren't the same thing.

I've split my content strategy into two tracks:

1. Traditional SEO pages: Long-form, comprehensive, link-building focus

2. GEO-optimized pages: Shorter, structured, answer-first, data-rich

They can be the same page, but only if you're intentional about structure. Most "SEO-optimized" content I see is actually terrible for AI extraction — too much fluff, buried leads, vague claims.

The pages that win in AI search aren't necessarily better written. They're better structured for extraction.

Most content teams are optimizing for a search engine that's increasingly mediating results through AI. If your content can't be easily extracted and cited, you're invisible to a growing share of your audience.

I'm not saying this is good or bad. I'm saying it's happening, and the data from my 30-day experiment makes that pretty clear.