Works
On-Page
Converting YouTube videos to written blogs
Written transcripts of video content get cited.
Aug 5, 2026
Read on otterly.ai →
Open research
GEO experiments and studies
AI search theories, tested and published openly.
OtterlyAI runs controlled experiments on how AI search engines like ChatGPT, Perplexity, Google AI Overviews and Copilot choose what to cite. We publish the full method and the raw data every time. If you want to check our work, you can.
20M+
URLs analyzed
7
AI search engines tracked
50,000+
Marketers using OtterlyAI
30
GEO Experiments + Studies Published
What's been tested
What works in AI Search?
Every row below comes from an experiment we ran and published, and each one links to the full write up with the method and the data.
What works in AI Search right now
- ✓Using AI instead of human written content
- ✓Placing key facts in your website's footer
- ✓Replying to Reddit threads
- ✓Use long form YouTube videos (YouTube Citation Study)
- ✓List your brand on directory websites
- ✓Get your brand into Top 10 listicles blogs
- ✓Add schema markup to your website
- ✓Publishing AI generated videos on YouTube
- ✓Clean URL structure (URL Citation Study)
- ✓Get mentioned on Analyst Relations sites (Analyst Relations Study)
- ✓Winning industry awards
- ✓Using LinkedIn Pulse articles instead of Posts (LinkedIn AI Citation Study)
- ✓Actively maintaining a Reddit community 15 min a day
- ✓Updating outdated pages and PDFs (Dead Citation Experiment)
- ✓Converting YouTube videos to written blogs
- ✓Pitching news publishers that have AI licensing deals
- ✓Recycling Livestreams into YouTube Videos + Blogs
What doesn't
- ✕Adding llms.txt to your website
- ✕Use Markdown (.md) pages instead of HTML
- ✕Hidden text as white on white background
- ✕Inject hidden prompt instructions on page
- ✕Optimizing image filenames, alt text and captions
- ✕Improving your content's design (AI vs Human Taste Study)
- ✕Publishing thousands of AI generated blogs with one click
- ✕Publish a Glossary Hub for Industry Terms
Still inconclusive
- ○Adding the year in your title and H1Edited pages grew 56% and 61%, but so did an untouched control.
Every result links to the full experiment, including the method and the raw data.
Open research
GEO Experiments
Controlled tests, published in full whether they worked or not.
Doesn't work
On-Page
Publishing thousands of AI generated blogs with one click
AI slop is fatal in Google. Both sites were silently deindexed.
Jul 31, 2026
Read on otterly.ai →
Works
On-Page
Updating outdated pages and PDFs (Dead Citation Experiment)
35% of a removed page's citations still arrived after deletion.
Jul 10, 2026
Read on otterly.ai →
Inconclusive
On-Page
Adding the year in your title and H1
Edited pages grew 56% and 61%, but so did an untouched control.
Jul 3, 2026
Read on otterly.ai →
Works
Social
Actively maintaining a Reddit community 15 min a day
Active subreddits earned 9x more AI citations than dormant ones.
Jun 15, 2026
Read on otterly.ai →
Doesn't work
On-Page
Improving your content's design (AI vs Human Taste Study)
AI has no shared taste. Design and aesthetics gave no benefit.
Jun 8, 2026
Read on otterly.ai →
Works
Off-Page
Winning industry awards
Effective but conditional. Award pages lifted visibility 121%.
May 23, 2026
Read on otterly.ai →
Works
Video
Publishing AI generated videos on YouTube
Exact match AI generated videos on YouTube get picked up.
May 1, 2026
Read on otterly.ai →
Doesn't work
On-Page
Optimizing image filenames, alt text and captions
No measurable impact. Image metadata alone informs nothing.
Apr 17, 2026
Read on otterly.ai →
Doesn't work
Black Hat
Injecting hidden prompt instructions on a page
No impact, and flagged as abuse by AI search platforms.
Apr 7, 2026
Read on otterly.ai →
Doesn't work
Black Hat
Hidden text as white on white background
Some platforms detect and discount it outright.
Apr 7, 2026
Read on otterly.ai →
Doesn't work
Technical
Using Markdown (.md) pages instead of HTML
No measurable impact on AI traffic or citations.
Apr 1, 2026
Read on otterly.ai →
Works
Technical
Adding schema markup to your website
Effective for Google AI Mode and AI Overviews only.
Mar 7, 2026
Read on otterly.ai →
Works
Off-Page
Getting your brand into Top 10 listicle blogs
Listicle mentions correlate strongly with being recommended.
Mar 3, 2026
Read on otterly.ai →
Works
Off-Page
Listing your brand on directory websites
Moderate impact. Certain niche directories carry real weight.
Mar 3, 2026
Read on otterly.ai →
Works
Social
Replying to Reddit threads
Effectiveness varies with the competitive intensity of the topic.
Mar 3, 2026
Read on otterly.ai →
Works
On-Page
Placing key facts in your website's footer
The fastest way to get a clear product statement picked up.
Mar 3, 2026
Read on otterly.ai →
Works
On-Page
Using AI instead of human written content
Optimized AI content performed 40x better than human content.
Feb 26, 2026
Read on otterly.ai →
Doesn't work
Technical
Adding llms.txt to your website
No measurable impact on AI traffic or citations.
Feb 5, 2026
Read on otterly.ai →
AI Search Studies
Large scale research into how AI search picks what it cites.
Featured study
ChatGPT
AI Licensing Deals Study
Publishers with OpenAI deals get 48% more citations per cited page.
Aug 15, 2026
Read the study →
Benchmarks
The 100 Most Visible Brands in AI Search
Company owned pages outweigh journalism four to one across 17 US industries.
Q3 2026
Read the study →
Social
Reddit Citation Study
ChatGPT cut reddit.com from 4.01% of its citations to 0.44%, an 89% drop in two steps.
Aug 2026
Read the study →
Where our research got picked up
- MarketingfactsYouTube GEO study: 94% of AI citations go to long-form videoRead more →
- Business InsiderFirst large-scale study shows YouTube is #2 social platform for AI citationsRead more →
- Search Engine LandHow query fan-out tools are used to optimize content and track resultsRead more →
- GlobeNewswireWhat licensing deals mean for publisher visibilityRead more →
Research methodology
Every experiment changes one variable at a time against a control, and the success metric is fixed before any data is collected. Where AI search cannot be randomised we use quasi experimental designs that apply the same one variable logic across matched groups.
One variable at a time
Only one thing changes between versions. If two change, the result cannot be attributed to either.
Define success before you start
The primary metric is locked before data collection begins and does not change once results arrive.
Run all versions at the same time
Parallel execution removes timing effects. A platform update hits every version equally.
Lock the experiment once it starts
No changes to format, posting time or treatment after day one. Mid-run changes invalidate the data.
Document decisions before looking at results
Analysis decisions are pre-registered. Anything decided after seeing data is labelled exploratory.
Set a minimum effect size
Results must clear a practical threshold to become a recommendation, not just be detectable.
About OtterlyAI
OtterlyAI is an AI search visibility platform. We track how brands appear in answers from ChatGPT, Perplexity, Google AI Overviews and Copilot. More than 50,000 marketing professionals use it to see where they show up and where they do not.
7
AI engines tracked
50,000+
Marketers on board
20M+
URLs analyzed
100%
Open methodology
Ready to see how AI search sees your brand?
Start tracking your visibility across the engines your buyers ask.
Try OtterlyAI