I audited 50 pages for AI search last week. Here's what actually moved the needle.

Last Tuesday, I was staring at our Search Console data, watching our "AI Overview" impressions climb while our click-through rate flatlined. The traffic was there, but nobody was clicking through. That's when I decided to stop guessing and actually audit what these AI engines were doing with our content.

I pulled 50 of our top-performing pages and ran them through a manual audit. Not the kind where you check meta tags and call it a day — I mean actually seeing how ChatGPT, Perplexity, and Gemini were parsing and citing our stuff.

Here's what I found, and more importantly, what I changed.

The problem with "SEO content" in an AI world

Most of our pages were written for Google's crawler. They had the right keywords, decent structure, and backlinks. But when I asked Perplexity to summarize our best-performing guide, it hallucinated three statistics and missed our core argument entirely.

The issue? Our content was structured for skimming humans, not for AI extraction.

If you're still thinking about this as just "SEO with extra steps," you need to read up on GEO vs SEO — the citation mechanics are fundamentally different.

What I actually tested

I took those 50 pages and ran them through three specific tests:

Test 1: The "First 200 Words" Check

I copied the first 200 words of each page and asked ChatGPT: "What is this page about?" If it couldn't give me a clear, accurate summary in one sentence, the intro was failing.

Result: 34 out of 50 pages failed this test. The intros were either too clever, too vague, or buried the lede under three paragraphs of context.

Test 2: The "Stat Verification" Drill

I pulled every statistic and claim from each page and asked Perplexity to verify them. This is where it got ugly.

Result: 12 pages had stats that AI engines either couldn't verify or actively contradicted. One page cited a "2023 study" that, when I checked, was actually from 2021. AI engines are getting better at catching this, and they're penalizing us for it.

Test 3: The "Direct Answer" Challenge

I took the main question each page was trying to answer and asked Gemini: "Answer this question directly." Then I compared its answer to our page's conclusion.

Result: Only 8 pages had conclusions that matched what AI considered the "correct" answer. The rest were either too nuanced (AI hates nuance) or too salesy.

What I changed (and the numbers)

Based on these tests, I made three specific changes:

1. I rewrote all 34 failing intros using the "BLUF" method (Bottom Line Up Front).

Instead of "In today's fast-paced digital landscape..." I went with "X tool increased conversion rates by 23% in our test. Here's how."

Early result: Pages with BLUF intros saw a 40% increase in AI citation rate over two weeks.

2. I added explicit "Key Takeaway" boxes to every page.

Not for humans — for AI. These are 2-3 sentence summaries at the top that state exactly what the page proves and how.

Example structure:

> Key Takeaway: [Main finding] because [evidence]. This matters for [audience] who need to [action].

3. I killed the fluff.

I cut an average of 300 words per page. Anything that didn't directly support the main claim got removed or moved to a "Background" section that AI clearly identifies as secondary.

The tool that saved me hours

I didn't do this entirely manually. I used a GEO Audit Tool to run a baseline scan on all 50 pages first. It flagged the structural issues (missing clear conclusions, buried key stats, ambiguous intros) so I could focus my manual effort on the content problems.

I also ran the pages through an AI Gravity Checker to see which ones had the authority signals AI engines actually care about. Turns out, our "About Us" page was outranking our product pages for brand queries because it had better entity markers. Who knew.

What's next

I'm now running a similar audit on our competitor's pages. The goal is to see what structure they're using that we're not. Early finding: our top three competitors all use a "TL;DR" box at the top of their posts. We don't. That's an easy fix.

I'm also testing whether adding schema markup specifically designed for AI (not just traditional SEO schema) makes a difference. More on that next week.

---

If you're still optimizing just for Google's crawler, you're leaving traffic on the table. The engines are changing, and our content structure needs to change with them.