The tests only matter if you know how good the evidence is. A new review of 9,249 SEO, AEO and GEO documents (The SEO Meta Study, Krešimir Ćorluka, Ovisy, Sept 10, 2026) graded every tactic it found. Below are the strangest results, and how much weight each one can carry.
Key takeaways
- Odd title tag changes produced some of the biggest measured swings, both up and down.
- Plain on-page work (titles, internal links, links) has the strongest evidence. AI-search tactics have the thinnest.
- Most published SEO results are uncontrolled and lean positive: 602 of 769 tactics have no controlled test, and 81.9% of published outcomes are wins.
- Treat every result here as something to test on your own pages, one change at a time.
How to read the grades: Controlled = changed pages compared with similar unchanged pages. Single practitioner = one person’s tests. Uncontrolled = before and after only. Bundle = several changes measured as one number.
Why this matters now
Everyone is rewriting their strategy for AI search. The review found 20 entries on title tags and 18 on internal linking, but only 10 on AI visibility. The industry talks most about the area it has tested least.
Title tag testing: the results
- ALL CAPS titles: +17.5% organic traffic (two controlled tests; +14% on travel listing pages). Strange, measured, and worth a test on one template.
- Trimming synonyms to shorten titles: -27% (controlled). The biggest single-test loss in the review. Long, descriptive titles keep earning their keep.
- Dropping the word “compare”: +24% (controlled). One wrong-intent word can cost clicks.
- Price in the title: -15% (controlled). On location templates with static rental prices: -7%.
- Emojis: negative every time they’re tested alone (number emojis in meta descriptions -5%; emojis in titles -6%). Bundled into a full redesign, impressions rose, but that number belongs to the redesign.
The link results
- Linking to orphaned blog posts: up to +64 positions (spread -4 to +64; a couple of pages did most of it).
- CTA blocks with internal links on product pages: +80% conversions (two controlled tests).
- Homepage footer links to ~100 category pages: +5% to +10% (one controlled test; in the public write-up, the overall 5% missed 95% confidence while desktop’s 10% cleared it).
- Digital PR links: +71% organic sessions (19 samples, none controlled).
- Disavow: +80% with an active manual action, no direction without one. No penalty, no disavow.
The AI-search results
- Direct answer in the first paragraph: +280% citations (single practitioner).
- Short factual statements instead of prose: 5x (single practitioner). Statistics added: +37%; named expert quotes: +22% to +41% (academic, controlled).
- Author bylines: +113% AI citations (7 samples) and human editing of AI text: +67% (5 samples).
- llms.txt: no controlled test at all. Three uncontrolled measurements swung from losses to +81%. Google says Google Search ignores the file (Search Engine Roundtable, June 16, 2026).
- AI text across a whole site: -35% sessions. Targeted AI copy on chosen pages: +73% citations.
Bonus from local: moving a Google Business Profile to a specific primary category gained 3.4 positions; going back to a broad one lost 30.
What the data says about SEO experimentation overall
Half the entries with big numbers (92 of 182) changed several things at once, so their number can’t be pinned on a single tactic. And 65 of the review’s 118 controlled measurements came from one publisher. The honest conclusion: the boring fundamentals are the best-tested part of SEO, and nearly everything else is a hypothesis.
Why most SEO experiments never get run
They copy a result from a screenshot, change ten things in the same month, and can’t tell which change helped. The fix is a controlled test on one template at a time. We walk through it step by step in our guide to SEO split testing.
SEO experiments for small business: three to start with
Small sites rarely have enough pages for a true split test, so run SEO experiments for small business as dated before and after tests:
- Rewrite the title tags on the five pages with the most impressions and the lowest click rate.
- Add internal links from your strongest pages to one service page.
- Add a short FAQ to one service page.
Log the date of each change and compare the same 28 days before and after in Search Console. Want to see real client results? Browse our SEO case studies.
FAQ
What are SEO experiments?
SEO experiments are deliberate changes to a website, such as rewriting title tags or adding internal links, measured to see whether organic traffic, rankings or conversions change.
Which SEO experiments have the strongest evidence?
In The SEO Meta Study, title tag changes, internal links and earned links have the most and best-controlled evidence. AI-search tactics have the least.
Is SEO experimentation worth it for a small business?
Yes, if you keep it simple: one change, on one group of similar pages, measured in Search Console against a fixed before period. Avoid changing several things at once.
Do llms.txt files improve AI visibility?
No controlled test shows they do, and Google says Google Search ignores them.
Sources
- Ćorluka, K. (2026). The SEO Meta Study. Ovisy, Sept 10, 2026, https://ovisy.hr/blog/seo-meta-study.
- SearchPilot, “Should you add more internal links to your footer for SEO?” (Sept 14, 2021).
- Search Engine Roundtable, June 16, 2026 (llms.txt).
Why we trust tests over case studies
Here’s the reality. I’ve built, scaled, and sold three companies, generated over $85 million in the last seven years in my own companies. We’ve watched plenty of tactics that looked great in someone else’s case study do nothing on our own pages. The study shows why: 81.9% of published SEO outcomes are wins and only 3.8% are losses, which tells you people publish what worked and bury the rest. A test on your own pages is the only result you can bank.
Want us to run your first test?
We pick one template, set up a control and a variant, change one thing, and read the result with you in Search Console. You get a written log of what won and what lost, and you approve every change before it ships. Not a fit if you want a one time trick. Why now: every month you copy a case study instead of testing is a month of guessing. See our on-page SEO services.
Want to know what actually works on your site?
Book a call and we’ll pick the first test worth running on your pages.