Check an AI statistic against the original data
Trace a fictional AI statistic to Eurostat’s original table. Check the year, population, geography and measure before you use the number.

In short
- Open the original dataset and check who, where and what the number measures, as well as its year and unit.
- Keep a percentage only when those details support your sentence. Rewrite the scope or leave the claim out if they do not.
- In this exercise, Eurostat's 19.95% applies to covered EU enterprises in 2025 using at least one measured AI technology. It cannot support a claim about every business worldwide using generative AI.
You can follow this exercise in a browser without an account or a paid AI tool. The sentence below is fictional practice text chosen for this guide. We checked the public dataset and its controls; we did not test a chatbot's accuracy.
1. Open the original source behind the number
Start with this practice claim:
"One in five businesses worldwide uses generative AI."
Open Eurostat's Use of artificial intelligence in enterprises (opens in a new tab). Its 2025 result is 19.95%, which rounds to about one in five. Follow its isoc_eb_ai dataset link, or open the Artificial intelligence by size class of enterprise table (opens in a new tab).
For your own claim, follow the citation through to the organisation that collected or published the data. A roundup that links to another roundup has not yet given you the table, survey or report needed to check the sentence. If you cannot find that original source, leave the claim unverified.
Check the dataset title and its data code, isoc_eb_ai, before changing any selections. The table's last update describes the dataset update; the Time dimension gives the observation year. Keep those dates separate.
Watch out
A familiar percentage can come from a different population or measure. Match the source's definitions before you copy the number. This Eurostat example covers selected EU industries and enterprises with at least 10 people employed, including employees and self-employed people. Survey scope (opens in a new tab)
2. Select one year and one population
Click Customise your dataset. For dimensions with several selected values, click Unselect all, then choose the value below. Wait for the selection count to update before switching dimensions.
- Under Time, choose 2025.
- Under Geopolitical entity (reporting), choose European Union - 27 countries (from 2020), code EU27_2020.
- Under Size classes in number of persons employed, choose 10 persons employed or more, code GE10.
- Under Information society indicator, choose the entry beginning Enterprises using at least one of the AI technologies, code E_AI_TANY.
- Check Unit of measure is Percentage of enterprises, code PC_ENT. Other units in the list calculate percentages for different groups.
- Keep Time frequency as Annual and the economic-activity selection C10-S951_X_K. The selection should contain one cell. Click Save and go to data view.
The same one-cell view (opens in a new tab) is available as a shortcut. It remains a live view, so it may reflect later data revisions.
The AI indicator includes technologies such as text mining, speech recognition, machine learning and autonomous movement, as well as generative technologies. You cannot rename this combined indicator "generative AI." Eurostat definitions (opens in a new tab)
3. Read the cell with its labels
Read the 2025 column and the EU row. The checked view shows 19.95, with Percentage of enterprises selected. Record it as 19.95%, not 19.95 enterprises.
The screenshot uses Eurostat's shorter geography code to keep the row readable. To use the same display, open Display settings, choose Labels and codes, scroll to geo, select Codes, clear Labels, then click Apply. You can keep the full country-group label while checking your own table.
The sector restriction still applies. Eurostat's metadata covers NACE sections C to N except K, plus group 95.1. That includes manufacturing, construction, trade and several service industries; it excludes agriculture and financial activities. Use "enterprises in the sectors covered by the survey" unless you have checked that your intended industry is included. The broad wording of the activity label in the table is not a reason to call the result "all businesses." Coverage and methodology (opens in a new tab)
4. Rewrite the sentence and save the evidence
A sentence supported by this selection is:
"In 2025, 19.95% of EU enterprises in the sectors covered by Eurostat's survey used at least one of the AI technologies it measured. The figure covers enterprises with at least 10 people employed, including employees and self-employed people."
Keep the dataset (opens in a new tab) and sector definitions (opens in a new tab) beside that sentence in your notes. "About one in five" is a reasonable rounded form of 19.95%, provided you keep the same year, population and technology definition.
The fictional claim broadens the result to businesses worldwide, drops the size and sector limits, and narrows the technology to generative AI. Eurostat's figure does not support those changes. Finding evidence for that broader claim would require another source check.
Save the data code, selected dimensions, exact cell value and date you accessed the source. A screenshot helps you retain the view, but keep a link and the definitions too. A live bookmark can change when the publisher revises the data.
If the table will not load or the numbers differ
If you see a browser error or an unfinished table, try one normal reload. Our custom view recovered after a browser rendering error. If it still will not load, use the source article only for the facts it actually states and mark the current table check incomplete. Do not claim you reproduced a cell you could not see.
If the article and table disagree, check their reference years, filters and update information. Eurostat warns that database updates can make current values differ from an article's earlier extract. Keep both figures and their contexts in your notes until you understand the difference.
Avoid extending this exercise into a growth calculation without checking comparability. Eurostat added an image, video and audio generation category to the 2025 questionnaire. Its detailed article also notes a 2025 break in the Netherlands series. Check those details before comparing years. Questionnaire change (opens in a new tab), series-break note (opens in a new tab)
Sources: Eurostat dataset isoc_eb_ai (opens in a new tab); custom selection (opens in a new tab); ICT usage in enterprises metadata (opens in a new tab); Use of artificial intelligence in enterprises (opens in a new tab); Eurostat reuse notice (opens in a new tab). Screenshots are cropped. The scope diagram and interpretation are adaptations by HoldMyTwist; Eurostat is not responsible for them.
Share this guide
From holdmytwist.com/guides/check-an-ai-statistic-original-data/


