Our tool gave us a perfect GEO score. Our content was still full of unsourced claims.
We built Foglift to help sites optimize for AI search engines — ChatGPT, Perplexity, Google AI Overviews. It scans your site and flags technical gaps: missing schema markup, robots.txt issues, structured data problems.
So we pointed it at ourselves. The automated scan came back with SEO 100, GEO 100, AEO 88. Three performance warnings. That's it.
Then we actually read our blog posts. What we found was embarrassing.
The problem automated scans don't catch
Our 9 pillar blog posts — the ones driving most of our organic traffic — were full of the exact patterns we tell our users to avoid:
- Unsourced statistics. Our GEO vs SEO comparison claimed "60-70% overlap between Google and ChatGPT results." No source. We'd picked that number up from a vendor blog that also had no source.
- Outdated data presented as current. We cited Google's daily search volume as 8.5 billion. The 2026 DemandSage data puts it at 13.7 billion+.
- Recycled vendor marketing claims. The widely-shared "44% increase in AI citations from schema markup" stat from BrightEdge? We repeated it. But the actual BrightEdge article doesn't contain that number — it describes structured data improving AI feature inclusion, not a 44% citation lift. We had never checked the primary source.
No automated scanner catches this. You have to read the content.
What we actually found in 9 blog posts
We went through every pillar post and cataloged the problems. Here's the pattern:
| Issue | Count across 9 posts | Example |
|---|---|---|
| Stats with no source | 14 | "68% of companies haven't started GEO" |
| Outdated numbers | 6 | Google searches: 8.5B → actually 13.7B+ |
| Vendor claims presented as research | 5 | "44% schema citation lift" (unsourced) |
| Missing Sources section | 9 of 9 | Zero posts had a references section |
| "Studies show" with no study named | 8 | "Studies show AI prefers structured data" |
Every single post had at least two of these problems. Our most-shared post — How ChatGPT Ranks Websites — had five unsourced claims in the first 500 words.
The fix: honest evidence over marketing claims
We spent 7 sessions upgrading all 9 posts. The process for each was the same:
- Find primary research. For every claim, we tracked down the original study — not the blog post that cited it, not the infographic that summarized it. The actual paper or report with methodology and sample size.
- Replace or remove unsourced claims. If we couldn't find a primary source, we either removed the stat or flagged it as unconfirmed.
- Add honest evidence callouts. Where vendor claims were exaggerated or unverifiable, we said so explicitly — including for popular stats we'd previously repeated ourselves.
- Add a Sources & Further Reading section. Every post now cites 6-12 original sources with author, title, year, and sample size where available.
Here's what replaced the unsourced claims:
| Before (unsourced) | After (cited) |
|---|---|
| "60-70% overlap between Google and ChatGPT" | Chatoptic study: 62% URL overlap, but only 0.034 rank correlation (1,000 queries, 15 brands) |
| "68% haven't started GEO" | Incremys 2026: 34% of companies have trained teams in GEO; 63% of marketers prioritize generative search |
| "44% citation lift from schema" | Google & Microsoft confirmed schema for AI features (March 2025); ChatGPT/Perplexity/Anthropic have not confirmed. Actual empirical data is mixed. |
| "Studies show freshness matters" | Seer Interactive: 71% of ChatGPT citations come from 2023-2025 content. Digital Bloom: updating within 30 days = 3.2x more citations. |
| "AI search is growing fast" | McKinsey (Aug 2025, 1,927 consumers): 44% prefer AI search. Bain 2025: 80% of search users rely on AI summaries ≥40% of the time. |
The schema markup post became our template for this approach. We wrote an explicit callout box: "What the research actually shows (2024-2026)" — separating what's confirmed by Google/Microsoft, what's not confirmed by ChatGPT/Perplexity, and what the empirical data actually says. Nuance that no other vendor blog in this space bothers with.
What we learned that surprised us
1. Most "GEO stats" trace back to 2-3 vendor reports, heavily paraphrased. We found the same BrightEdge and Gartner numbers recycled across dozens of blogs, each time with slightly different framing and less context. The telephone game makes every stat less accurate.
2. The largest study is barely cited. SE Ranking and Search Engine Journal analyzed 129,000 domains and 216,524 pages across 20 niches — the biggest ChatGPT citation study to date. Key findings: expert quotes increase citations from 2.4 to 4.1; 19+ data points increase citations from 2.8 to 5.4; referring domains are the single strongest predictor. We found this study referenced in maybe 5% of the "how to optimize for AI" articles we surveyed.
3. Honest uncertainty earns more trust than false confidence. Saying "this isn't confirmed" is more valuable than confidently citing an unverifiable stat. The schema markup post that explicitly calls out what's unconfirmed has become one of our most-linked pieces.
4. The research bar is low. Adding primary sources to a blog post in this space immediately puts you in the top 10% of content quality. Most vendor blogs in GEO/AEO cite each other, not the research. The bar for being the honest-evidence source is surprisingly achievable.
What this means for developers building content
If you're writing technical content for SEO or AI search visibility:
- Trace every stat to its primary source. If you can't find the original study with methodology and sample size, flag it as unverified or drop it.
- Add a Sources section. Academic-style — author/org, title, year. This is table stakes in research but rare in tech content.
- Say "not confirmed" when it's not confirmed. AI companies are opaque about ranking factors. Honesty about uncertainty is a competitive advantage.
- Cite the actual sample sizes. "A study of 129,000 domains" carries more weight with both humans and AI models than "research shows."
- Update stats every 3-6 months. We had 2024 numbers presented as current on a page dated 2026. AI models weight freshness heavily — 71% of ChatGPT citations come from content published 2023-2025.
- Run your own tools on yourself. The automated scan was a useful starting point. The real value came from the manual content audit it prompted.
Where we stand now
After the 9-post upgrade, our site scans at Overall 95, SEO 100, GEO 100, AEO 88. The AEO gap is performance-related (server response time), not content.
Every blog post now has 6-12 cited sources. Zero unsourced "studies show" claims remain.
Foglift is free — if you want to run the same scan on your own site, it takes about 10 seconds. But the real audit starts when you read your own content with fresh eyes and ask: "Where did this number actually come from?"
That's the question that fixed our content. It'll probably fix yours too.
Built by the Foglift team. We scan for GEO + AEO readiness so AI search engines actually cite you. Currently scoring ourselves at 95 — the remaining 5 points are a performance problem we're still working on.
Top comments (0)