Anyone can claim they saw AI search coming. The test is whether they wrote it down first. This page is where I do that. Each prediction has the date I made it, a date it gets checked, and the evidence it will be checked against. On the check date I mark it right, wrong or couldn’t score, and the row stays on the board either way. Missed calls don't get deleted or quietly reworded. The first call was checked on 6 November 2026. It was wrong, and it’s still on the board.
The ledger
When each call gets checked
Nov 2026 this month 3
Dec 2026 2
Jan 2027 8
Mar 2027 1
Apr 2027 3
Oct 2027 7
Dec 2027 8
- Right
- 0
- Wrong
- 1
- Still open
- 31
- Couldn’t score
- 0
24 hours after the announcement, at least one of the four engines will give a stale answer in at least one of its two runs.
- Written down
- Checked
- Counts as right if
- Bank Rate catch-up study, 24-hour round on 6 November 2026. Hit if one or more of the eight 24-hour answers is stale. Miss if none is, or the round isn't run and published by 13 November 2026. Void if the decision isn't announced on 5 November 2026.
- Outcome
- Scored 6 Nov 2026: miss. None of the eight 24-hour answers (run 12:02 to 12:07 UK) was stale; six were caught up and two unreferenced. Data ↓
After the follow-up 'When is that figure from? Give the month or period and the official source.', at least 95% of AI answers will state a period for the figure.
- Written down
- To be checked
- Counts as right if
- 76 or more of 80 included arm D second answers state a period. Miss if 75 or fewer, or not run and published by 30 Nov 2026.
The reasoning, where it was first published →
Note: Baseline: 33 of 40 first answers dated on 10 Oct 2026.
In the same run, at least one period an engine gives after the date follow-up will not match the official period.
- Written down
- To be checked
- Counts as right if
- At least one arm D period differs from the official period read the same day. Miss if all match, or not run and published by 30 Nov 2026.
The reasoning, where it was first published →
Note: Periods were not scored for correctness on 10 Oct 2026.
On the first Russell Group robots.txt snapshot, a smaller share of the 24 universities will block GPTBot than the UK's ten biggest newsbrands did (7 of 10).
- Written down
- To be checked
- Counts as right if
- GPTBot blocked by 16 or fewer of 24 on the 2 Dec 2026 snapshot. Miss if 17 or more, or not published by 9 Dec 2026.
The reasoning, where it was first published →
Note: Baseline: news 7 of 10 on 2 Oct 2026; universities not measured.
On the first Russell Group snapshot, universities will block training crawlers more often than retrieval crawlers, by at least 10 percentage points.
- Written down
- To be checked
- Counts as right if
- Training cells blocked / 216 at least 10 points above retrieval cells blocked / 144. Miss if smaller, or not published by 9 Dec 2026.
The reasoning, where it was first published →
Note: Baseline: news 74% vs 53% on 2 Oct 2026.
On a rerun of the same ten UK newsbrands and 15 crawlers, at least 7 of 10 will still block training crawlers more often than user-triggered fetchers.
- Written down
- To be checked
- Counts as right if
- At least 7 of 10 sites block more of {GPTBot, ClaudeBot, CCBot, Google-Extended} than of {ChatGPT-User, Claude-User, Perplexity-User} on the 2 Jan 2027 snapshot.
Google-Extended will remain the least-blocked training token in the same ten-site sample (5 of 10 on 2 Oct 2026).
- Written down
- To be checked
- Counts as right if
- No other training-class bot is blocked by fewer sites than Google-Extended on the 2 Jan 2027 snapshot.
On the first findings recall audit, no more than 3 of 72 AI answers will credit jacobwhitmore.com by link or name.
- Written down
- To be checked
- Counts as right if
- 3 or fewer of 72 answers scored linked or named. Miss if 4 or more, or not published by 9 Jan 2027.
The reasoning, where it was first published →
Note: Baseline: none.
On the same audit, at least one AI answer will state one of the nine findings in full without crediting jacobwhitmore.com.
- Written down
- To be checked
- Counts as right if
- At least one row with finding present = full and credit = none. Miss if none, or not published by 9 Jan 2027.
The reasoning, where it was first published →
Note: Baseline: none.
On the first run of the registered link, mention or nothing study, fewer than half of scored answers will be linked.
- Written down
- To be checked
- Counts as right if
- Linked rows are fewer than half of all scored rows. Miss if half or more, or if the study has not been run and published by the check date.
The reasoning, where it was first published →
Note: 2026-10-10: published as right/wrong/inconclusive; scored as hit/miss/void with the same tests.
On the same first run, every engine in the study will give at least one mention-only answer.
- Written down
- To be checked
- Counts as right if
- Each engine has at least one mention-only row. Miss if any engine has zero, or if the study has not been run and published by the check date.
The reasoning, where it was first published →
Note: 2026-10-10: as P3.
At least three of the four AI search statistics already replaced by their own source (Ahrefs 34.5%, SparkToro 58.5%, OpenAI 900 million weekly users, Gemini 950 million monthly users) will still appear on three or more statistics roundups on the 29 January 2027 rerun.
- Written down
- To be checked
- Counts as right if
- Rerun with the same six searches, first ten results each, same roundup definition. Hit if at least 3 of the 4 appear on 3+ roundups. Miss if 2 or fewer. Void if fewer than 15 roundups open.
The reasoning, where it was first published →
Note: Baseline 29 Oct 2026: all 4 on 3+ roundups (34.5%: 6; 900M: 7; 950M: 3; 58.5%: 3).
None of the AI search statistics repeated on three or more roundups on 29 January 2027 will be measured in the UK.
- Written down
- To be checked
- Counts as right if
- Same rerun. Hit if no statistic on 3+ roundups has the UK as its stated market (a multi-country figure with a UK breakdown counts as UK). Miss otherwise. Void if fewer than 15 roundups open.
The reasoning, where it was first published →
Note: Baseline 29 Oct 2026: 0 of 28 traced statistics were UK.
The CMA will impose its user choice conduct requirement on Google by 31 March 2027.
- Written down
- To be checked
- Counts as right if
- The CMA publishes a final decision imposing the user choice conduct requirement on or before 31 Mar 2027. Miss if later or dropped.
The reasoning, where it was first published →
Note: Confidence 80% set on 17 Oct 2026, never changed. Called on /predictions/seo-ai-search-2027/. Baseline: consultation closed 9 Oct 2026; the CMA expects a final decision by the end of 2026. Evidence: https://www.gov.uk/government/news/cma-strengthens-proposals-allowing-people-choice-over-their-search-service
Meta's Web Crawlers page will not list a separate user-agent token for Muse's own browser.
- Written down
- To be checked
- Counts as right if
- Right if on 2 April 2027 Meta's Web Crawlers page names no token that mentions Muse or describes the Muse Secure VM browser. Wrong if it does.
The reasoning, where it was first published →
Note: Called on 23 Oct 2026 on the Muse post. Added to this CSV on 29 Oct 2026; it was on the board page from 23 Oct but missing from the 17 Oct CSV.
No more than 3 of the 10 UK newsbrands in the crawler sample will publish an llms.txt file at the domain root on 2 April 2027.
- Written down
- To be checked
- Counts as right if
- Same ten domains, each read in a browser on the day. Hit if 0 to 3 return a readable llms.txt. Miss if 4 or more do.
The reasoning, where it was first published →
Note: Baseline 30 Oct 2026: 1 of 10 (Daily Mail).
Fewer than 4 of the 13 AI visibility trackers checked on 30 October 2026 will publish accuracy or validation evidence for their own numbers on 30 April 2027.
- Written down
- To be checked
- Counts as right if
- Same 13 tools, same nine questions, each vendor's own public pages read on the day. Hit if 0 to 3 publish validation evidence. Miss if 4 or more do.
The reasoning, where it was first published →
Note: Baseline 30 Oct 2026: 0 of 13.
A year after the first snapshot, at least nine of the same ten UK newsbrands will still block at least one of the same 15 AI crawlers.
- Written down
- To be checked
- Counts as right if
- 9 or 10 of 10 sites block at least one crawler on the 2 Oct 2027 snapshot. Baseline 9 of 10.
The reasoning, where it was first published →
Note: 2026-10-10: as P3.
Squarespace's "Block known artificial intelligence crawlers" setting will still be off by default.
- Written down
- To be checked
- Counts as right if
- The setting is off by default on a new Squarespace site on the check date.
The reasoning, where it was first published →
Note: 2026-10-10: given ID P6.
At least 2 of ChatGPT, Gemini and Perplexity will still name a single "best plumber in Leeds".
- Written down
- To be checked
- Counts as right if
- At least 2 of the 3 engines name one specific company for the site prompt on the check date.
The reasoning, where it was first published →
Note: 2026-10-10: given ID P7.
At least 3 of 5 engines will quote a Semrush UK price with no date attached.
- Written down
- To be checked
- Counts as right if
- At least 3 of 5 engines give a price with no date for the site prompt on the check date.
The reasoning, where it was first published →
Note: 2026-10-10: given ID P8.
Google's spam policies will still not treat AI authorship on its own as a violation or require an AI label for Search eligibility.
- Written down
- To be checked
- Counts as right if
- developers.google.com/search/docs/essentials/spam-policies on the check date has neither. Baseline: page last updated 28 Aug 2026.
The reasoning, where it was first published →
Note: 2026-10-10: given ID P9.
A year from now, AI engines will still quote the purple pound without the year it refers to.
- Written down
- To be checked
- Counts as right if
- Same prompt on ChatGPT, Gemini, Google AI Mode, Perplexity and Claude, twice each, UK, one day: of the engines that answer, at least half give a figure with no year. Baseline 2 of 2 answering.
The reasoning, where it was first published →
Note: 2026-10-10: published as right/wrong; scored as hit/miss/void with the same test.
A rerun will show the same persona split: at least 3 of 4 finance-director answers recommend none of the products in that day's no-persona answers.
- Written down
- To be checked
- Counts as right if
- 3 or 4 of 4 finance-director answers on ChatGPT and Gemini, twice each. Baseline 3 of 4.
The reasoning, where it was first published →
Note: 2026-10-10: given ID P11.
A signed-out search on google.co.uk will still land on the standard results page, not an AI Mode thread.
- Written down
- To be checked
- Counts as right if
- A query typed into google.co.uk, signed out, in a fresh Chrome profile from a UK connection, returns the standard results page. Miss if it opens AI Mode by default.
The reasoning, where it was first published →
Note: Confidence 70% set on 17 Oct 2026, never changed. Called on /predictions/seo-ai-search-2027/. Baseline: Google rebuilt the Search box on 19 May 2026 but said people would keep getting a range of results; AI Mode was not made the default. Evidence: https://blog.google/products-and-platforms/products/search/search-io-2026/
Google won't move the AI Overviews and AI Mode opt-out into robots.txt. The control will still be a Search Console setting.
- Written down
- To be checked
- Counts as right if
- Google's documentation names no robots.txt user-agent or token that removes a site from AI Overviews or AI Mode while keeping it in Search on the check date.
The reasoning, where it was first published →
Note: Confidence 75% set on 17 Oct 2026, never changed. Called on /predictions/seo-ai-search-2027/. Baseline: the only opt-out is a Search Console toggle; Google-Extended does not cover AI Overviews or AI Mode. Evidence: https://blog.google/products-and-platforms/products/search/new-controls-website-owners/
Google's guide to generative AI search will still say Google Search ignores llms.txt.
- Written down
- To be checked
- Counts as right if
- The guide still says Google Search doesn't use llms.txt or similar files on the check date. Miss if the line is removed or reversed.
The reasoning, where it was first published →
Note: Confidence 85% set on 17 Oct 2026, never changed. Called on /predictions/seo-ai-search-2027/. Baseline: guide last updated 10 Jul 2026 says llms.txt neither helps nor harms because Google Search ignores it. Evidence: https://developers.google.com/search/docs/fundamentals/ai-optimization-guide
The UK will be added to the countries where Google shows ads inside AI Overviews.
- Written down
- To be checked
- Counts as right if
- Google Ads Help 'About ads and AI Overviews' lists the United Kingdom for ads in AI Overviews on the check date. Miss if not listed.
The reasoning, where it was first published →
Note: Confidence 55% set on 17 Oct 2026, never changed. Called on /predictions/seo-ai-search-2027/. Baseline: 12 countries listed on 17 Oct 2026; the UK is not one. Evidence: https://support.google.com/google-ads/answer/16297775
ChatGPT Plus will still be ad-free.
- Written down
- To be checked
- Counts as right if
- OpenAI's 'Ads in ChatGPT' help page still says Plus accounts will not have ads on the check date. Miss if Plus gets ads.
The reasoning, where it was first published →
Note: Confidence 80% set on 17 Oct 2026, never changed. Called on /predictions/seo-ai-search-2027/. Baseline: ads on Free and Go only; Plus, Pro, Business, Enterprise and Edu ad-free, read 17 Oct 2026. Evidence: https://help.openai.com/en/articles/20001047
Search Console will report clicks, not only impressions, from Google's generative AI features.
- Written down
- To be checked
- Counts as right if
- Any Search Console report or filter shows clicks attributed to AI Overviews or AI Mode on the check date. Miss if impressions only.
The reasoning, where it was first published →
Note: Confidence 65% set on 17 Oct 2026, never changed. Called on /predictions/seo-ai-search-2027/. Baseline: impressions and pages only, rolled out worldwide 31 Aug 2026; Google says more metrics will follow. The CMA publisher conduct requirement asks for engagement metrics: https://www.gov.uk/find-digital-markets-measures/google-search-publisher-conduct-requirement Evidence: https://blog.google/products-and-platforms/products/search/new-controls-website-owners/
Google's UK search share on StatCounter (all platforms) for November 2027 will be at least 88.00%.
- Written down
- To be checked
- Counts as right if
- The November 2027 figure on StatCounter's UK all-platforms page is 88.00% or higher. Miss if lower.
The reasoning, where it was first published →
Note: Confidence 85% set on 17 Oct 2026, never changed. Called on /predictions/seo-ai-search-2027/. Baseline: 91.39% for September 2026, read 17 Oct 2026. Evidence: https://gs.statcounter.com/search-engine-market-share/all/united-kingdom
At least one AI assistant will be live on Google's UK search choice screen on Android or Chrome.
- Written down
- To be checked
- Counts as right if
- A CMA or Google publication confirms an AI assistant listed on the live UK choice screen by the check date. Miss if proposed only.
The reasoning, where it was first published →
Note: Confidence 35% set on 17 Oct 2026, never changed. Called on /predictions/seo-ai-search-2027/. Baseline: proposed on 23 Sep 2026; no AI assistant on Google's choice screens. Evidence: https://www.gov.uk/government/news/cma-strengthens-proposals-allowing-people-choice-over-their-search-service
How a row gets marked
On its check date I rerun the stated check and mark the row hit (the test is met in full) or miss (it isn't; the row stays up, in red). A call is void only if the thing being measured stops existing, such as a site closing or a crawler token being retired, not because the result is awkward. Every row links to its evidence. The full test for each call is on its source page.
Correction log
| Date | IDs | What changed |
|---|---|---|
| P1–P11 | Nine calls made in posts were added to the board and given IDs P3–P11. Status labels changed from right / wrong / inconclusive to hit / miss / void. No claim or test was reworded. | |
| P12–P22 | Six calls added from that day’s three study plans, as P17–P22. P12–P16 are reserved for calls that need the five-engines baselines. They have not been made, so they are not on the board. No existing call was changed. | |
| P23–P31 | Nine new calls added from the 2027 page, with confidence figures. No existing call was changed, and none got a confidence figure after the fact. | |
| P32–P34 | P33 and P34 added from the AI search statistics study. P32 was on this page from 23 Oct but missing from the board CSV; the 29 Oct CSV includes it. No existing call was changed. | |
| P35–P36 | Two calls added from the AI visibility trackers study and the llms.txt study. No existing call was changed. | |
| P37 | The Bank Rate catch-up call, made on the plan page on 11 Oct 2026, added to the board. The plan page called it P19, which was already the date follow-up call, so it is now P37. Claim and test unchanged. | |
| P37 | Scored 6 Nov 2026: miss. In the 24-hour round (12:02 to 12:07 UK) none of the eight answers was stale: six named the 5 November decision and two gave 4% without naming it. Results and partial data (CSV). Claim unchanged. | |
| 8 Nov 2026 | P1–P37 | The ledger is now generated from the board CSV, so this page and the homepage can’t disagree. Statuses are shown as Right, Wrong, Still open and Couldn’t score (hit, miss, open and void in the CSV). No claim or test was changed. |