AI SEO Research
Every public study I have published on how AI search reads, cites and sends traffic: 24 pieces across six first-party datasets, each dated, linked and labelled by how much weight it can carry. Studies land here first; the long-form reading of what they add up to is on the AI SEO studies write-up.
Updated 29 September 2026. New reads are appended with their date; null results stay on the page.
Datasets
Six datasets carry all 24 studies. 8 of the studies sit on StudioHawk or Search Engine Land because the agency owns the data; the rest are on this site. A study that shares a dataset with another is the same evidence read from a different angle, so count the datasets before you count the articles.
The 100-brand Australian ecommerce dataset
Last read 20 September 2026 · measured
100 Australian ecommerce brands, 21 months. GA4, channels split by referrer. July 2024 to March 2026.
ChatGPT sent 340,000 sessions and about $690,000 of revenue; Google still drives 96.2% of search revenue; Bing Organic converts at 3.6%.
9 studies
- utm_source=chatgpt.com: What It Means and What the Traffic Is Worth
- Search Engine Differences Explained
- Perplexity Referral Traffic in GA4
- What Your Board Is Asking About AI Search (And the Honest Answer)
- What 100 Australian eCommerce brands taught us about AI search (The High-Ticket Effect)
- Gemini Referral Traffic in GA4
- AI Search and Ecommerce Stats (2025)
- AI Search and Ecommerce: The 100-Brand Report
- Chosen, Not Clicked: eCommerce SEO Guide to Visibility in 2026 (white paper)
The 1.2 million AI sessions format dataset
Last read 24 August 2026 · measured
1,203,748 AI referral sessions across 600+ businesses, 7.5 million pages classified. GA4 landing pages classified by format. 15 January to 14 July 2026.
Interactive tools and calculators earn 7.5 times their share of pages; comparison pages rank fifth of seven formats.
3 studies
The 1,000+ audits dataset
Last read 22 August 2026 · measured
1,000+ StudioHawk SEO and AI search audits across seven verticals, plus an 83-site llms.txt check. Audit records. Trailing twelve months to mid 2026.
89% weak titles and meta, 84% URL problems, 84% content gaps, 81% weak E-E-A-T; 82% of 83 top Australian sites have no llms.txt.
2 studies
This site as the lab
Last read 21 September 2026 · measured, September bot mix directional
lawrencehitches.com, about 380 live pages, 108 finished controlled tests since 2023. Search Console, SEOtesting, Bing Webmaster Tools, Cloudflare verified-bot categories, GA4, Clarity. March to September 2026 for the three-channel reads.
Median test outcome 0.0%; one page lost 93% of Google clicks while AI cited it more; 17,211 citations across 313 grounding queries.
6 studies
- Claude Referral: What claude.ai/referral Means in GA4
- AI Search ROI: How to Measure the Return
- Visual Semantics and Topical Authority: I Ran the Test
- How AI Is Changing SEO (and Whether It Is Killing It)
- How I Ranked for AI SEO Consultant Melbourne in AI Overviews, ChatGPT, Gemini and Claude
- I Lost 93% of My Clicks. AI Cites Me More Than Ever.
Attribution experiments
Last read 11 February 2026 · directional
Four field experiments across client and own-brand accounts. CRM attribution and prompt tracking. 2025.
AI search compresses consideration, not discovery: more than $100,000 of closed revenue from 20+ AI-influenced leads in a year.
1 study
Method
- Every change is annotated the day it ships, in SEOtesting, and material changes become tests with a control period. That is what makes a single-page read like the EntityMap result possible.
- Three channels, three instruments. Google clicks from Search Console, Bing clicks from Bing Webmaster Tools, AI grounding from Cloudflare verified-bot categories and the Bing AI Search report. Impressions are never read as demand.
- Agency datasets are GA4 exports across the StudioHawk client estate, landing pages classified by format and channels split by referrer. Per-brand figures stay anonymous.
- One vendor per table. Keyword difficulty and volume are proprietary scales, so no table mixes vendors.
- Nulls are published. The Mushroom definitional treatment read -1% against a -1% control and was retired. The visual topical layer test read flat on five pages and was retired.
Confidence labels. Measured means a controlled read on real data with a named instrument. Directional means a real number from a sample too small to bank, or a metric that moves week to week. Nothing on this page is an estimate.
Applying the research
- Split your channels first. If Google clicks are falling while position holds, you are in the pattern the clicks-to-citations study describes, and the fix is on the Bing and citation side, not the title tag.
- Audit format before copy. Compute your own AI Traffic Leverage Ratio; if tools and calculators over-earn on your site the way they did across 600 businesses, build one before the next blog post.
- Map what you are already cited for. Run a prompt set through the AI search tracker and the brand visibility check; defend share where you have it, win more where volume is high and share is low.
- Research queries as fan-outs. Query fan-out is the research unit for AI engines; the free keyword tool labels intent and funnel stage on a whole list at once.
- Test one change against a control and read it on the date. The 108-test archive says the median change does nothing, so the read is the product, not the change.
Working on this for a business? Start with the AI SEO consulting page, or take the free AI SEO roadmap.
The social and video search dataset
Last read 31 July 2026 · directional, one channel
StudioHawk's YouTube channel in Google Search, first weeks of platform properties. Search Console platform properties. July 2026.
18,233 Google impressions on day one, 152,000+ by day eight, 1,005 queries in two days with 303 on page one.
3 studies