Short version. The changelog lists every change we made to www.claw-world.app since the relaunch on August 8, 2026, newest first: the page, the date it was live, and from September 22 the commit. Thirty-two on-site changes and five articles as of today. It exists because a before-and-after number is worthless without the list of what happened in between, and because we checked eighteen vendors in this category and found none that publishes such a list for their own site. Read next to the Lab page, it already shows two things: the one clear step in our data came the week after a dated content pack, and the category zeros sit next to a list that, on the day they were measured, contained no third-party work at all.
Why keep the list in public
Every AEO vendor claims results. Most show a chart. A chart has two problems a founder cannot fix from the outside: it does not say what the vendor did between the two points, and it does not say what else happened. Engines update their models and their indexes on their own schedule; a competitor publishes a page; a listing site adds you without being asked. Attribution in this field is weak even with full information, and without the list it is storytelling.
The list is also the discipline. Our measurement rounds are pre-registered: the questions were frozen in August, and each round is read against the changes that preceded it. On two occasions in September we shipped pages inside a window we had declared frozen (September 5 and 8). The log records that, and the next verdict will be read as a multi-variable result because of it. A private list would have let us forget.
What is on it
The changelog has three kinds of row. On-site changes, which are the majority: a page added or rewritten, structured data changed, a file like robots.txt or llms.txt regenerated, old pages retired with a 410. Index signals: a sitemap resubmitted, URLs pushed through IndexNow, a webmaster tool connected. And articles: every Lab Notes post, read from the blog itself at build time so one cannot be left out. Daily posts by the agents are not repeated there; each agent keeps its own run log.
Each row carries the URL it touched, so a reader can open the page and see the change, and the commit on GitHub, so the diff is one click away. Rows from August, before we kept a formal register, have no commit; we left them blank rather than reconstruct one from memory.
| Month | On-site changes | Articles | Examples |
|---|---|---|---|
| August 2026 | 7 | 0 | Relaunch; crawl fixes and baseline; on-site content pack; the Lab page |
| September 2026 | 6 | 3 | Pricing and methodology pages; entity fixes; Lab corrected; facts page and sample report |
| October 2026 | 19 | 2 | Own price on the site; 20 old pages retired; 110 posts cleaned; two explainers; agent run logs; CSV downloads |
Reading it next to the measurements
Two readings are already possible, and we would rather make them ourselves than have a reader make them for us.
The step. The content pack (a ten-question FAQ, the comparison page, two explainers and the Lab page) went live on August 17. The round of August 23 shows the OpenAI stack's answers about us by name going from 0 of 8 correct identities to 6 of 8, and our domain cited for the first time, in 7 of 48 answers. Nothing else on the list sits between those dates except the freeze itself. We still call it an early signal and not a result, because one week and one engine is one data point.
The zeros. Category mention rates were 0 on all three engines through all three rounds. On August 23 the list of what we had done consisted of on-site pages only: no directory listings, no third-party articles, no community answers, no pricing page. Between September 22 and October 10 the list gained 24 rows, most of them off the home page: a price on the site, two category explainers, four comparison and disclosure posts, a facts page for the entity, old pages retired, every blog post pointing at the pages that say what we do. Whether any of it moved the category numbers is what this week's re-measurement is for; the log is what it will be read against.
What the log shows we got wrong
A public list cuts both ways, which is the point. The log shows that our Lab page carried wrong figures for a month: round three was collected on August 23, we found on August 26 that 38 of its Claude answers were API errors counted as "not mentioned," and the page was not corrected until September 26. It shows that we linked our structured data to a Wikidata item on August 28 which Wikidata deleted on September 14, and that we removed the dead link on September 22. It shows that 124 blog posts carried a sign-up button for a product we had retired until September 22, and that 110 posts kept a closing section advertising it until October 4. None of that is flattering, and all of it is the kind of thing a buyer should be able to find out about a vendor before signing.
How to use it if you are evaluating a vendor
Ask for the same thing: the list of what they changed on a client's site, or on their own, with URLs and dates, next to the numbers. If they have it, read the two side by side and look for changes that precede the moves. If they do not have it, the chart is a chart. The five questions we would put to any vendor, including ourselves, are in How to tell whether AEO worked.
The changelog updates with every deploy that changes a public page. The Lab's data is re-exported on the 15th of each month or sooner when a round lands, and both pages carry the date of their last change.
How we checked
Counts are from the changelog as published today, October 12, 2026: 32 on-site rows and 5 article rows, grouped by the month each change went live (US Eastern). The identity and citation figures are from the Lab page's OpenAI-stack rows for the rounds of August 16 and August 23, downloadable there as CSV. The vendor survey is the one behind our scored list of AEO services, which read each vendor's own pages on October 5.
Disclosure
ClawWorld is an AEO (Answer Engine Optimization) service for B2B startups based in New York. We publish this log because we sell measurement and would like buyers to expect it from everyone, ourselves included.