1. The single source
| Document | Publisher and date | Retrieval |
|---|---|---|
| Philip R. Lane, “AI adoption in Europe”, keynote at the Federal Reserve Bank of San Francisco Economic Research Symposium on AI and the economy | European Central Bank, 30 October 2026 | Retrieved directly on 31 October 2026; HTTP 200; 211,807 bytes raw HTML; read in full |
URL: https://www.ecb.europa.eu/press/key/date/2026/html/ecb.sp261030~08d9d540f0.en.html
2. Claim table
Each row is a claim made in the article, the exact string tested, and the result of an exact-string search of the retrieved page text after normalising markup, entities and whitespace.
| # | Claim in article | String tested | Result |
|---|---|---|---|
| 1 | AI added about 10 per cent to capex plans this year | “added about 10 per cent to their capital expenditure plans” | FOUND |
| 2 | AI has been the main driver of global goods trade | “main driver of global goods trade” | FOUND |
| 3 | Data-centre investment is a relatively small share | “relatively small share” | FOUND |
| 4 | 64 per cent report experimental or moderate use | “64 per cent” | FOUND |
| 5 | Only 7 per cent report significant use | “only 7 per cent” | FOUND |
| 6 | Digital investment up 73 per cent in real terms since 2014 | “expanded by 73 per cent in real terms since 2014” | FOUND |
| 7 | Digital investment rose from 2.3 to 3.3 per cent of GDP | “from 2.3 per cent to 3.3 per cent of GDP” | FOUND |
| 8 | 62 per cent cited improving business processes | “62 per cent cited improving business processes” | FOUND |
| 9 | 23 per cent cited innovation, R&D or product range | “23 per cent cited innovation” | FOUND |
| 10 | 15 per cent cited reducing labour costs | “15 per cent cited reducing labour costs” | FOUND |
| 11 | Peer misperception of up to 50 percentage points | “as much as 50 percentage points” | FOUND |
| 12 | Firms underestimate peer adoption | “firms underestimate the rate of AI adoption by their peer firms” | FOUND |
| 13 | Source of the 10 per cent figure | “Corporate Telephone Survey” | FOUND (4 occurrences) |
| 14 | Source of the 64 / 7 per cent figures | “Survey on the Access to Finance of Enterprises” | FOUND (7 occurrences) |
| 15 | Survey round is Q4 2025 | “The latest observations are for the fourth quarter of 2025” | FOUND |
| 16 | Lending to AI-related firms has contributed substantially to credit growth | “lending to AI-related firms has contributed substantially to aggregate credit growth” | FOUND |
Absence control (deliberate)
Secondary coverage of this speech attributes to it a claim that AI-related lending is roughly one-quarter of recent credit growth. The string “one-quarter” does not appear anywhere in the speech. Because the absence is what the check found, the article does not report the figure, and states in its transparency note why. This is a control in the strict sense: a search that returned it would have proved the search unsound.
3. Quotation harness
A second, independent pass extracted every quoted span from the draft and tested each against the retrieved page text. Results:
| Measure | Result |
|---|---|
| Quotation spans tested | 9 |
| Matched exactly | 8 |
| Matched after one disclosed substitution (inner double quotation marks rendered as single, British nesting convention) | 1 |
| Matched on capitalisation only | 0 |
| Not found | 0 |
The substitution is disclosed in the article immediately under the blockquote it affects. No word of any quotation was altered, added or removed. Two spans in the draft marked by quotation marks are the reporter’s own phrasing, not attributions to the source (“AI replaces workers” and the secondary-coverage figure named in the absence control); they are excluded from the count above.
4. Separating survey findings from conclusions
The article’s core editorial requirement was to keep what the surveys measured distinct from what was inferred. The mapping used:
| Statement | Status |
|---|---|
| Firms report AI added about 10 per cent to capex plans | Survey self-report, as quoted |
| 64 per cent use AI experimentally or moderately; 7 per cent significantly | Survey self-report, as quoted |
| Most firms underestimate peer adoption, some by 50 percentage points | Survey result, from a field experiment within the survey round |
| Lending to AI-related firms contributed substantially to credit growth | ECB’s own analytical statement, quoted |
| Europe is financing AI rather than using it | Our framing of the gap between the above. Presented in the headline and standfirst as a summary, with the survey basis stated in the article body and the limits listed. |
| AI is not raising European productivity | Not claimed. Listed under what the numbers do not establish. |
5. What this file does not prove
- That the ECB’s surveys are accurate. The file proves the article reports them faithfully, not that they are right.
- The SAFE sample size, fieldwork dates or the definition of “significant” use, none of which appear in the speech.
- Anything about the underlying survey datasets, which were not retrieved.
6. Status
PUBLISHED October 31, 2026. Approved by the human publisher. This file publishes with the article, as every THAF story does.