Free resource
An SEO checklist where every item shows its evidence
I spent the summer reading 9,249 documents to find out which SEO tactics anyone has actually measured. Then I put all of them in a spreadsheet and graded the evidence behind each one.
Most SEO checklists are a list of things someone believes. This one is a list of things someone measured, sorted by how good the measurement was. Where the measurement is bad, it says so. Where there is no measurement at all, there is a whole tab about that.
Krešimir Ćorluka
CEO @ Ovisy · 28 September 2026
What the grades mean
Every row gets a letter from A to X. The letter rates the evidence, not the tactic. A well measured change can still be wrong for your site.
- A
Somebody isolated a single change, ran a control group, and at least two independent publishers found the same thing. There are 13 of these out of 161.
- B
The same thing with one publisher behind it. 28 of the 36 B rows are SearchPilot, which I will come back to.
- C
Several publishers agree, but nobody ran a control group.
- D
The study changed four things at once and reported one number.
- E
Weak. One or two publishers, no control, or a source that is not really a study.
- X
Do not act on this. Deprecated mechanics, or evidence that falls apart the moment you open the source.
The study reports 182 tactics and 552 primary sources. This sheet grades the 161 of those tactics that carry a traceable source list, and counts all 757 source rows behind them, which is why the numbers differ.
For the full methodology and source review, read The SEO Meta Study.
The thing that surprised me most
The size of the number tells you about the study, not about the tactic.
+16% is the median result of a controlled A/B test.
+106% is the median result of a before and after with no control group.
Same industry. Same tactics. Six times the claim, measured across all 757 sources.
So if a row in this sheet promises you +300%, look at its design column before you believe it. That is also true of every case study you will read this week.
Three tabs you will not find anywhere else
Backfired
Eight changes that were measured to make things worse. Removing synonyms from listing title tags cost 27% of organic sessions. Appending the price to a product title cost 15%. Removing the last self-referential breadcrumb link cost 5.5%.
All eight come from controlled tests, and seven were published by the same company, because almost nobody else in this industry publishes a loss.
Never tested
Twenty things we all recommend and nobody has ever isolated. XML sitemaps, hreflang, crawl budget, nofollow attributes and HTTPS migration do not appear anywhere in 757 sources. Page speed, URL structure, robots.txt, canonical tags and the primary category on a Google Business Profile do appear, but never once separated from four other changes made at the same time.
That list is on every technical audit template in this industry, including mine. That is the finding. We recommend those things on consensus, because isolating them is hard, and nobody has done it.
Withdrawn
Twelve rows I pulled out, with the reason next to each one. A headline sourced from a services page. A vendor PDF with no date. Two tactics Google killed while I was writing.
How to use it
Open the Prioritise tab and filter the Area column to whatever you are working on. Read the evidence grade before you read the reported effect. Check the precondition and the warning flag before you act on anything.
Then tick column A as you go. If you make your own copy in Google Sheets, select column A and use Insert, then Tick box, and you get real checkboxes.
The priority score is evidence times replication divided by effort. The size of the effect is deliberately not in it, because in this corpus a bigger number mostly means a weaker design, and ranking on it would put the worst evidence on top. All the weights are editable on the Scoring tab. Move them and everything recalculates.
What this cannot tell you
Whether a row applies to your site. Nothing here is segmented by site size, traffic band, CMS or market. A 23% internal linking result on a 40,000 SKU retailer is not a forecast for a 12 page B2B site.
What order to do them in. The score ranks evidence. Sequencing depends on what your last release broke and what your developers can ship this month.
What happens when you do several at once. Every A and B row isolated one change. Nobody has measured the interaction between them.
Whether the effect lasts. Half the entries do not state a measurement window at all.
The limitation I will say out loud first
SearchPilot carries 55 of the 85 controlled tests in the whole corpus. The next publisher has five.
Their tests are the best public evidence that exists in this field, and they run mostly on large ecommerce and travel sites. Before you generalise a row to your own site, check what it was tested on.
I report that as a limitation of my own work. It is also the finding: almost nobody else runs controlled tests and publishes the results, including the ones that failed.
It can still be wrong
Every row traces back to a source you can open in one click. If one of them is wrong, tell me and I will publish the correction with your name on it.
How to cite it
Ćorluka, K. (2026). The SEO Checklist, graded by evidence, version 1.0. Built from The SEO Meta Study. https://doi.org/10.5281/zenodo.22768160
The checklist is CC BY 4.0. Copy it, build on it, sell work around it. Just leave the line at the bottom of each tab.
Why it is free
Because a finding you have to pay for is a finding nobody can check.
The study is free, the checklist is free, and both will stay that way. Applying them to a specific site is the work I sell. That part has always been the actual job.
About Krešimir Ćorluka
Krešimir is CEO and co-founder of Ovisy. He has worked on hundreds of websites across SEO, technical SEO, content, digital PR and search strategy.
Built from The SEO Meta Study, a systematic review of 9,249 SEO, AEO and GEO documents. Entry numbers in the checklist match the entry numbers in the study.