Optimizely
Optimizely is an experimentation platform. Sites use it to run split tests: showing one version of a page to some visitors and a different version to others, then measuring which performs better. This index classifies it as a tracking service under A/B testing, because deciding which visitor sees which variant, and remembering that decision on the next page, requires identifying the visitor. It was detected on 69 of the 5,294 sites measured, 1.3 percent, ranking 58th of the 88 companies here by reach.
Sixty-seven of sixty-nine
On 67 of the 69 sites, at least one company was contacted before the banner window closed. That is 97.1 percent, and it ranks 15th of the 88 companies in this index. Median first contact came at 596 milliseconds, 63rd of 88, so outside activity tends to begin earlier on these pages than on most sets measured here. The quickest recorded across the group was 146 milliseconds, and bbc.co.uk matches that figure exactly. Split testing works by choosing a variant before the page paints, which is why these scripts are conventionally placed high in the document — but that is a fact about how the technique works, not something this crawl measured. What the crawl records is how many companies a site contacted before the window closed, not which ones.
Half of them run OneTrust
45 of the 69 sites carried a recognisable consent tool, and OneTrust was seen on 35 of them: half of the entire set, not just half of those with a banner. The IAB TCF signal appeared on 7, and Iubenda, Cookiebot and Osano on 1 each. Those are sighting counts rather than slices, since one site can run more than one layer. A single consent vendor covering half a set says something about who the customers are. OneTrust sells into organisations large enough to have a compliance function, and the site list matches: broadcasters, a hospital system, developer-tool companies and a workplace messaging platform.
Unremarkable weight
The median Optimizely site contacted 7 companies, 60th of 88; made 41 third-party requests, 50th of 88; and carried 2 trackers, 44th of 88. Median companies contacted before the window closed was 6, 37th of 88 and level with 19 others, and at the median no company is contacted after it. Nothing here is extreme in either direction. The profile is a mainstream commercial site with a moderate stack, not a publisher carrying an ad exchange, and that is worth stating because a tool classified as tracking does not necessarily sit on heavy pages.
An engineering signature in the co-presence list
Google was on 44 of the 69 sites, 63.8 percent, and OneTrust on 35, 50.7. Then Google Tag Manager and unpkg on 11 each, 15.9 percent; jsDelivr, Sentry and Cloudflare Analytics on 9 each, 13; Microsoft Ads on 8, 11.6; Amazon CloudFront and tiqcdn.com on 7 each, 10.1. Sentry and unpkg appearing in the same band is a tell: error monitoring and a package delivery network are the marks of a site maintained by an engineering team with a build process, rather than one assembled in a page builder. That fits a product whose whole premise is shipping variants of a page deliberately.
The sixty-nine
bbc.com and bbc.co.uk both registered 6 companies with all 6 inside the window, and bbc.co.uk began earliest of the examples at 146 milliseconds. cnn.com registered 17 with 14 inside and fortune.com 15 with 14. typeform.com and klaviyo.com registered 14 each, clevelandclinic.org 11, slack.com 10, change.org 8 with 4 inside, gofundme.com and bit.ly 7 apiece, jetbrains.com 5. change.org was the latest of the examples to begin, at 1,045 milliseconds. Nonprofits and a hospital system sit in the same set as newsrooms and software companies, which is what an experimentation tool with a broad customer base looks like from the outside.
| Site | Companies | Before consent | First contact |
|---|---|---|---|
| bit.ly | 7 | 5 | 920 ms |
| cnn.com | 17 | 14 | 522 ms |
| bbc.com | 6 | 6 | 258 ms |
| bbc.co.uk | 6 | 6 | 146 ms |
| slack.com | 10 | 8 | 806 ms |
| klaviyo.com | 14 | 10 | 587 ms |
| change.org | 8 | 4 | 1045 ms |
| jetbrains.com | 5 | 5 | 474 ms |
| typeform.com | 14 | 13 | 410 ms |
| fortune.com | 15 | 14 | 162 ms |
| gofundme.com | 7 | 6 | 817 ms |
| clevelandclinic.org | 11 | 11 | 608 ms |
| Tool | Sites |
|---|---|
| OneTrust | 35 |
| IAB TCF consent tool | 7 |
| Iubenda | 1 |
| Cookiebot | 1 |
| Osano | 1 |
Seen on the same pages
Companies most often contacted by the same sites. Co-presence only — the crawl records what arrived, not who called whom.
- GoogleTogether on 44 of these sites
- OneTrustTogether on 35 of these sites
- Google Tag ManagerTogether on 11 of these sites
- unpkgTogether on 11 of these sites
- jsDelivrTogether on 9 of these sites
- SentryTogether on 9 of these sites
- Cloudflare AnalyticsTogether on 9 of these sites
- Microsoft AdsTogether on 8 of these sites
Questions
What does Optimizely do on a page?
It decides which version of the page a given visitor should see, applies that version, and reports back on how visitors in each group behaved. From the browser's point of view it is a script that runs early and a stream of measurement calls afterwards.
Why is an A/B testing tool classified as tracking here?
Because assigning a visitor to a variant and keeping them in it across pages requires an identifier tied to that visitor. This index lists Optimizely under A/B testing and marks it as a tracking service on that basis.
Did Optimizely load before the consent banner on these sites?
The crawl cannot answer that. It records that 67 of the 69 sites had some company contacted before the window closed, without recording which company it was on any given site. Presence on a site with early contact is not the same as being the cause of it.
How many Optimizely sites show a consent banner?
45 of the 69 carried a recognisable consent tool, and OneTrust was seen on 35 sites, which is more than three quarters of those with a banner and half of the set overall.