How to write a market research brief with sources you can check

A consultancy is considering an introductory AI workshop for small businesses. Before designing it, the team wants to know what it can say about potential buyers. An adoption statistic can help frame the investigation. It cannot establish that businesses will pay for training.

A market research brief with sources starts with a specific decision and gives the reviewer a way to check each factual claim. This example uses a fictional assignment and real public statistics about businesses in the European Union. The consultancy and starting draft sentence are synthetic demonstration inputs, not a client engagement or observed commercial results.

1. Define the decision the brief needs to support

State the question, target segment, geography and observation period before collecting evidence. A short brief should make clear which conclusions the available material supports and which questions still require investigation.

Our question is: what do official 2025 statistics establish about EU businesses with 10–49 people, and what remains unknown before offering them an introductory AI workshop? The survey's headcount includes employees and self-employed people.

The assignment calls for adoption context, a labelled commercial hypothesis and questions to ask potential buyers. It does not call for a revenue forecast or a count of businesses willing to book. Those conclusions would need evidence about needs, budgets and existing training options.

Set a practical acceptance rule: a reviewer must be able to open every source, find the supporting passage and identify the population behind the claim.

2. Read the original publication and its scope

Use the organisation that collected the data, and read the methodological notes alongside the headline. Keep the publication date separate from the period the data describes.

The example uses Eurostat's news release published on 11 December 2025 and its explanatory article using data extracted in December 2025. The release supplies a rounded overview; the article adds business size, reasons for non-adoption and definitions. Both come from the same organisation, so treat them as related presentations rather than independent corroboration.

The survey covers specified economic activities and businesses with at least ten people. It also covers several AI technologies. Carry those restrictions into the brief: the figures do not represent every European business or generative AI alone.

For a national or sector-specific assignment, find evidence that matches that population and check its scope again. An EU-wide result can provide context, but it cannot establish the needs of buyers in one town or industry.

3. Ask for a draft that keeps facts and inference separate

Give the model the question, source links and evidence fields you want returned. Require it to identify sources it cannot open and leave the associated claims unverified.

This reusable prompt condenses the assignment. Adapt the segment and source list to your own question:

1Write a short brief for a fictional consultancy considering an
2introductory AI workshop for EU businesses with 10–49 people.
3Question: what do official 2025 figures establish, and what must
4we still check before investing in the offer?
5
6Open these Eurostat sources:
7https://ec.europa.eu/eurostat/web/products-eurostat-news/w/ddn-20251211-2
8https://ec.europa.eu/eurostat/statistics-explained/index.php?title=Use_of_artificial_intelligence_in_enterprises
9
10Separate verified facts, inferences and unanswered questions.
11For every fact record the publisher, URL, publication or extraction
12date, reference period, population and supporting section. Describe
13the supporting passage in your own words and label it a paraphrase.
14
15If you cannot open a source, mark the claim as unverified.
16Do not invent interviews, willingness to pay or a market size.
17Produce a DOCX brief, the same brief as a PDF and a CSV claim ledger.

Choosing the initial sources remains part of the analyst's work. A brief may expose a substantial gap in the evidence. Keep that gap visible so the person making the decision knows what further work is needed.

4. Check every factual claim in a ledger

Review the claims separately from the narrative and record the passage supporting each one. A claim ledger helps a reviewer spot a correct percentage attached to the wrong population.

These are the core checks for the example. Every figure concerns the surveyed EU sectors. The passage descriptions are paraphrases, with links to the original pages.

In the working file, give each row an identifier that connects it to the relevant sentence. Update the row whenever you edit the claim, so the checked version and the delivered version remain aligned.

5. Correct the conclusion the source cannot support

A sentence may need both a narrower population and removal of an unsupported commercial conclusion. Keep the original and the correction together so the reviewer can see what changed.

We deliberately supplied an overbroad draft sentence in Spanish. Translated into English, it reads:

“70.9% of European businesses lack AI expertise; therefore, there is proven demand for AI courses.”

Checking Eurostat reveals two problems. The percentage belongs to a selected group of non-users, and the survey question concerns adoption barriers rather than training purchases.

Against the supplied HTML, the run corrected the population and removed the unsupported commercial conclusion. This historical passage from the original Spanish output translates as follows:

Among EU enterprises in the surveyed sectors with at least ten people employed that had considered AI but were not using it in 2025, 70.89% cited a lack of relevant expertise as a reason. This figure covers all surveyed size groups; it does not isolate businesses with 10–49 people or establish willingness to pay for a workshop.

Unresolved source discrepancy: the HTML used in the run reports 70.89%, while Table 7 of Eurostat's 2026 report gives 70.3% for the same all-size 2025 group. We have not resolved this difference. Do not reuse the sample's percentage as a settled estimate until it is resolved.

Commercial hypothesis to test: some businesses with 10–49 people would pay for an introductory AI workshop built around a task they already do. The cited statistics do not establish that demand. Ask potential buyers which task they want to improve, what support they need and whether they would pay for it.

6. Deliver the brief with its unanswered questions

Include the claim ledger and keep unresolved questions in the report itself. The person reading the recommendation should be able to see what still needs checking before committing resources.

For this assignment, useful questions include the work a business wants to improve, who would approve the training, the available budget and what the team has already tried. Answer them through conversations with potential buyers or other relevant evidence. An EU adoption percentage cannot supply those answers.

Open both the DOCX and PDF before sharing them. Check the figures and links, make sure the scope notes remain beside the conclusions, and inspect tables for clipped text. When using the example to teach the method, keep its fictional assignment clearly labelled.

We ran this example in Ilisai's test environment on 6 September 2026, in Spanish with the selector set to Auto. The original outputs preserved below contain the same 479-word report in DOCX and PDF, with Eurostat links, plus a CSV with five claims and eleven fields. The sample files are in Spanish across all three language versions of this guide.

These original QA outputs preserve what the run produced. Their percentage and ledger statuses reflect the supplied HTML comparison and have not been reconciled with Table 7. The discrepancy note appears only in this article; the files remain unchanged. Do not reuse the sample's percentage as a settled estimate until the discrepancy is resolved.

Download the original QA files:

• Original QA report (DOCX, Spanish)
• Original QA report (PDF, Spanish)
• Original QA claim and source ledger (CSV, Spanish)

It took three corrections after the first delivery. The initial DOCX applied an adoption barrier from the wider surveyed population to the 10–49-person target segment and omitted the requested PDF and CSV. The first revision improved the conclusion but returned an abridged PDF, an incomplete ledger and documents without functioning hyperlinks.

For the second correction, we supplied a 479-word master report and a complete ledger, both matched to the supplied Eurostat HTML pages. The resulting DOCX and CSV matched those inputs, with functioning document links and the required ledger fields; the PDF still omitted the source-date paragraph. One final request restored it. We then compared both documents' text and checked the links, page legibility and all five CSV rows. These checks established text and file consistency; they did not resolve the statistical discrepancy. The preserved files came from Ilisai after those instructions; we did not edit their content locally.

In the enterprise: assign the review

Name the person who checks the sources and the person who approves the recommendation when the brief will support a consequential decision. Specify the version they review and the changes that require another check.

A strategy or procurement team may require additional sources or buyer interviews before accepting a proposal. Put that requirement in the assignment and name the person who decides whether the evidence is sufficient to recommend the purchase.

Build a brief for a real business question

In Ilisai, you can work with web sources and generate a downloadable DOCX or PDF report; check the claims and the finished file before sharing it. The AI document generator explains the available formats, while our guide to generative AI for business places research alongside other practical workflows; see pricing for usage terms.