26 September 2026 · SEO Dispatch
Most SEO evidence is uncontrolled. How SEO Dispatch is built differently.
Why borrowed case-study percentages are a weak forecast, what the Ovisy SEO Meta Study found in the public corpus, and how SEO Dispatch measures your site instead.
Most SEO "proof" you see online is not a controlled test of one change on a site like yours. It is often a write-up after several edits shipped together, on a different template, in a different month of Google updates. That can still be useful direction. It is a weak forecast for your URL.
Industry skew and denominators
When someone quotes a big percentage lift, ask what sat in the denominator. How many pages. How long. Whether a holdout existed. Whether null results were published at all.
The SEO Meta Study (Ovisy, September 2026) is a systematic review with narrative synthesis of public SEO, AEO, and GEO case material, not a statistical meta-analysis. It analysed 9,249 documents. 1,391 of them measured something. 552 primary sources sit behind that measured set. The published book carries 182 tactic entries after evidence gates. The same review logged 769 clusters that passed those gates before the printed entry cut. That is a map of what got written up on the open web, not a promise that the same lever will move your traffic by the same amount. Full PDF: https://ovisy.hr/seo-meta-study.pdf.
Publication bias cuts the other way too: practitioners publish the tests that moved a number. Agreement across uncontrolled write-ups overstates confidence. Absence of a tactic in a review means absence of published measurement that met their gates, not proof the tactic never works.
Bundled changes
92 of 182 entries in that study (51%) rest on at least one source that changed more than one thing at once. The single percentage then belongs to the package, not to one tidy lever. Direction can still matter. Treating the size as a clean A/B for one checkbox does not.
Ovisy also keeps two scores apart: E (how strong the evidence is) and A (how worth doing the tactic is). A change can be well evidenced and still a poor use of your week, or the reverse. We will not collapse that into a single "proven" badge on SEO Dispatch.
What we refuse
- Selling industry case-study percentage gains as your expected lift.
- Calling AI answer / GEO tactics a sure substitute for crawlable, clear pages.
- Inventing clicks, CTR, position, or queries from Search Console's generative AI performance views (those views are impressions-oriented for AI Overviews / AI Mode).
- Treating Google's AI-optimization guidance as if it handed out ranking hacks. Google's May 2026 AI optimization guide frames generative AI features in Search as rooted in core ranking and quality systems. Special AI-only files or schema are not Search ranking requirements.
What we measure
On your property we aim to combine:
- Crawl evidence on real URLs (what bots can fetch and parse).
- Google Search Console signals that belong to this site (coverage, queries, pages; generative AI impressions only when that report applies). Generative AI performance reports are impressions by page. That report does not include clicks, CTR, position, or queries.
- Your Focus (what you sell and who you want to reach).
- Demand gates so open work means work worth doing on this site, not every raw scare signal.
Fix Guides stay human-reviewed. We never publish to your site. We recommend changes. We do not guarantee outcomes.
Credit
Figures and denominators above are from Krešimir Ćorluka, September 2026, The SEO Meta Study, Edition 1, Ovisy (https://ovisy.hr/seo-meta-study.pdf). We paraphrase for product honesty. We do not republish the study. Cite Ovisy and the original publishers when you reuse a measurement.
AI Overview citations are observational, not an RCT
Large crawls (for example Ahrefs work on AI Overview citations and CTR) are useful context. They are observational. Parser changes and SERP redesigns confound neat before/after stories. Ranking in the top results often helps citation odds. It does not guarantee an AI Overview cite. Big average CTR gaps when an Overview is present are not a law for every query on every site. Label the design. Do not turn a cohort estimate into a product promise.