The Jacob Whitmore Index is a fixed set of numbers about how UK websites and AI search treat each other. Each number comes from a study published on this site with its data file. Each one has the date it was measured and the date it will be measured again. If something hasn’t been measured yet, the Index says so and doesn’t guess.
It sits inside the UK AI Search Observatory, which holds the studies. The Index is the short version: the readings worth quoting, in one table.
Current readings
Last updated 15 November 2026. A reading changes only when a new dated study is published. The old reading stays in that study and in its CSV.
| # | Measure | Latest reading | Measured | Next reading | Study and data |
|---|---|---|---|---|---|
| JWI-1 | Cells blocked: the UK’s ten biggest newsbrands against 15 AI crawlers | 99 of 150 | 2 Nov 2026 | 2 Dec 2026 | UK news robots.txt, November 2026 |
| JWI-2 | UK newsbrands blocking at least one AI crawler | 9 of 10 | 2 Nov 2026 | 2 Dec 2026 | UK news robots.txt, November 2026 |
| JWI-3 | Cells blocked: the UK’s ten biggest supermarkets against the same 15 crawlers | 0 of 150 | 10 Nov 2026 | 10 Feb 2027 | UK supermarkets and AI crawlers |
| JWI-4 | Cells blocked: the UK’s ten most-visited retail sites against the same 15 crawlers | 27 of 150 | 12 Nov 2026 | 12 Feb 2027 | UK retail sites and AI crawlers |
| JWI-5 | AI company sites that block rival AI crawlers | 3 of 17 | 6 Nov 2026 | 6 Feb 2027 | AI companies and robots.txt |
| JWI-6 | UK news and Russell Group university sites that publish llms.txt (of those that could be read) | 5 of 33 | 30 Oct 2026 | 30 Jan 2027 | llms.txt on UK news and university sites |
| JWI-7 | AI answers to “current” UK figure questions that gave no date | 7 of 40 | 10 Oct 2026 | 10 Jan 2027 | Do AI answers say when a figure is from? |
| JWI-8 | Most-repeated AI search statistics that pass all five source checks | 0 of 32 | 29 Oct 2026 | 29 Jan 2027 | AI search statistics, traced to the source |
| JWI-9 | AI visibility trackers that publish evidence their numbers are accurate | 0 of 13 | 30 Oct 2026 | 30 Apr 2027 | What AI visibility trackers say about how they measure |
| JWI-10 | The UK’s ten most-visited retail sites named in at least one of 36 AI shopping answers (ChatGPT, Google AI Mode, Gemini, Claude) | 9 of 10 | 14 Nov 2026 | 14 Feb 2027 | Which UK retailers do AI tools name? |
JWI-1 to JWI-4 use the same 15 crawlers and the same definition of blocked, so they can be compared. What each crawler does is on AI crawlers explained. Every CSV is listed on the data page, and the figures can be recomputed from the scripts.
Not measured yet
These are the questions the Index is built to answer next. None has a published reading. Each line says what has to be collected first.
| Question | What has to be collected | Status |
|---|---|---|
| Which websites do AI tools link to when answering questions about search? | The same 50 fixed questions asked every month, with every linked host recorded | First reading taken 14 Nov 2026 on Google AI Mode. Not published |
| Which UK businesses do AI tools name and recommend? | Fixed prompts per sector, three runs each, on each AI tool, with every business named and linked | Retail pilot published 15 Nov 2026 (JWI-10). The other nine sectors have not been run. Plan: UK businesses in AI answers |
| Does AI visibility move over time? | A second and third reading of the two lines above, same prompts, same settings | Waiting on the first readings |
| Do the sites Google ranks match the sites AI tools cite? | Google’s top ten for each of the 50 questions, collected the same day as the AI answers | Not started |
| Which sectors get named most in AI answers? | The full ten-sector run of UK businesses in AI answers | Not started |
| Do AI tools correct out-of-date figures when asked? | The registered date follow-up run: 160 chats on four tools | Run 14 Nov 2026. Results page due by 30 Nov 2026 |
| Are businesses mentioned more on other sites named more in AI answers? | For each business in that ten-sector sample, a count of independent pages naming it, from a stated source on a stated date | Method not fixed. Will be registered before any count is taken |
What gets into the Index
- The number comes from a study published on this site, with its CSV.
- The sample is fixed and listed, so a rerun measures the same thing.
- It has a measured date and a next date. If a next date is missed, the row says so.
- A reading is never edited after publication. A new study adds a new reading. Mistakes are corrected in public under the editorial policy.
- Predictions made from a reading go on the ledger and are scored there, including the ones that miss.
What every Index study contains
Every study behind a reading uses the same layout, so readers know where to look:
- The research question, as a question.
- A key takeaway in one or two sentences, with the number in it.
- The date the data was collected.
- The sample: what was included, how it was chosen and its size.
- The method: exact prompts or files read, tool settings, and the definition used for each count.
- The results, in a table.
- What this doesn’t show.
- The data file (CSV, CC BY 4.0) and how to recompute it.
- Any prediction made from it, with its ledger number.
- The next reading date.
- A cite line.
Studies that ask AI tools questions are registered as a plan first, in the Search Experiment Lab, so the prompts can’t be changed after the answers come in. The full rules are on Method.
How to cite a reading
Quote the reading with its date and the Index number, and link the study. For example: “99 of 150 AI crawler checks were blocked across the UK’s ten biggest newsbrands on 2 November 2026 (Jacob Whitmore Index, JWI-1).” Free to use with a link. Questions or corrections: contact@jacobwhitmore.com. Journalists: see press.
Limits
- The samples are small and named: ten newsbrands, ten supermarkets, ten retail sites. They describe those sites, not the whole UK web.
- A robots.txt file is a request. A blocked cell says what a site asks for, not what a crawler does.
- AI answers change from run to run and by account. Readings from AI answers are counts from stated runs on stated dates, not rates.
- One person collects and codes the data. There is no second coder.