Guide

How to write a research report

A report earns citations when it contains a number nobody else has. Everything else — the design, the summary, the launch — only matters once that is true.

Neubook Write · 11 September 2026 · 8 min read

Start from a question people argue about

A report that answers a settled question gets read by nobody. Look for the places your industry disagrees without evidence: what people actually charge, how long something really takes, whether a practice everyone recommends is followed, what proportion of the work is unpaid. The best research questions are ones where practitioners have strong opinions and no data.

Write the headline finding you hope to publish, as a sentence, before you collect anything. If it would not make a practitioner stop scrolling, choose a different question.

Then commit to publishing whatever the answer turns out to be. A report designed to produce a predetermined conclusion is marketing dressed as research, and readers can tell.

Where the data comes from

The most defensible reports combine two: a survey for opinion and platform data or public records for behaviour. Where the two disagree, you have your most interesting section.

Designing a survey that does not mislead you

• Ask what people did, not what they would do. Hypothetical answers are close to worthless.
• One idea per question. "Do you find the process slow and expensive?" produces uninterpretable answers.
• Avoid leading phrasing. "How much has AI improved your workflow?" has already assumed the finding.
• Offer ranges for money and time, which raises the response rate and reduces made-up precision.
• Record who you asked and how. The recruitment method is a finding in itself.
• Keep it under five minutes. Completion rate collapses after that, and partial responses skew your data.

Collect only what you need and say how you will handle it. If you are collecting from people in the EU or UK, the GDPR basics apply: a lawful basis, a clear privacy note, and no keeping identifiable data longer than necessary. Anonymise before analysis wherever you can.

The structure of the report

1. Summary — one page, five findings as sentences with numbers in them. Written last.
2. Method — early, not hidden in an appendix. Who you asked, how many, when, how they were recruited, what you excluded. Putting this up front is the strongest credibility signal available to you.
3. Findings — one section per finding, each with the number, the chart, and two paragraphs of interpretation.
4. What surprised us — the section readers quote. Be honest about what contradicted your expectations.
5. Limitations — sample size, self-selection, geography, anything you could not measure. This section makes the report more credible, not less.
6. Appendix — the questionnaire, and the data tables if you can share them.

Charts that do not lie

• Start bar charts at zero. Truncated axes are the most common way honest people mislead.
• Label the n on every chart. "n = 62" next to a percentage prevents the most common misreading.
• Avoid pie charts beyond three segments; people cannot compare angles.
• Show distributions, not just averages. A median plus a range tells a truer story than a mean, especially for money.
• Design for reuse. Charts get screenshotted. Put your URL discreetly on each one.

Writing it up without overclaiming

Two disciplines separate a cited report from an embarrassing one.

Correlation language

"Respondents who did X also reported Y" is a finding. "X causes Y" is a claim you cannot support with a survey. Keep the verbs honest, and the report survives scrutiny by the one expert who reads it closely.

Every number traceable

If AI helps draft the prose, adopt a hard rule: no number enters the text except by being copied from your analysis. Models interpolate plausible figures with total confidence, and a single invented statistic is enough to destroy the credibility of the whole document. Check the draft against the dataset line by line before publishing.

Publishing it so it gets cited

• Put it on a permanent URL on your own domain, with a year in the title.
• Make the key numbers copyable as text, not trapped in images — journalists and LLMs both need text.
• Offer a citation line ("Source: Neubook Write Industry Report 2026") near the top. Make it trivially easy to credit you.
• Publish the questionnaire and, if possible, the anonymised data. This is rare, generous and disproportionately rewarded with links.
• Send it directly to the ten people in your industry who write about this. One personal email each beats a press release.
• Repeat it annually. The second edition is where the value compounds: year-on-year change is the most citable thing you will ever have.

Google's guidance on people-first content rewards exactly this kind of first-hand material — original data, clear method, demonstrable expertise — which is why an honest report outperforms a dozen commodity blog posts.

Turning it into a book

A report is also a publishable object: an EPUB and a print-ready PDF give you something to hand out at a conference, sell at a modest price, or send to a journalist who will not open a link. The book version wants a longer introduction, the charts redrawn for print, and the method section kept exactly as it is.

A four-week plan

Choosing between a survey and behavioural data

Most first reports use a survey because it is the easiest thing to run. It is also the weakest evidence, because people misremember and misreport. Where you can, pair it with something observed.

The gap between the two columns is frequently the most quoted finding in the whole report, because it is the thing nobody could have asserted without data.

Sample size, honestly

Independent researchers agonise over sample size and then overclaim anyway. A more useful frame:

• Under 30 responses: report as interviews, use quotes, avoid percentages entirely. "Nine of the twenty-two said…" is honest; "41 percent" from n=22 is not.
• 30–100: percentages with the n stated on every chart, no subgroup breakdowns beyond two or three categories.
• 100–500: the useful range for an industry report. Subgroups become possible if each has enough respondents.
• Above 500: you can start comparing segments with some confidence — and should say plainly how respondents were recruited, because self-selection does not disappear with scale.

Nobody has ever discredited a report for saying "this is a small sample from a self-selected group, and here is what it suggests". Reports get discredited for the opposite.

The summary page

Most readers read one page. Write it last, and build it from five findings stated as sentences containing numbers:

1. Freelancers in this survey raised prices once every 2.4 years;
   the median rise was 11%.
2. Those who raised prices annually reported 28% higher income
   than those who raised them "when it felt necessary".
3. Eight in ten had no written pricing policy of any kind.
4. The most common reason for not raising prices was fear of
   losing a specific named client, not market conditions.
5. Respondents with a written policy were twice as likely to
   have turned work down in the last year.
   (n = 214, recruited via three industry newsletters, March 2026)

Every line is quotable, carries its own number, and does not require the rest of the document. That is what makes a report travel.

The annual edition, which is the real asset

A single report is content; a repeated one is an institution. From the first edition onward:

• Keep the core questions identical year to year, even the ones you would now word better. Comparability is worth more than perfection.
• Add no more than three new questions per edition.
• Publish the change, not just the level. "Down four points from 2026" is the headline the second year earns you.
• Keep the raw anonymised data in a stable place, versioned.
• Name it consistently so citations accumulate to one title rather than scattering.

By the third edition you own a dataset nobody else has, which is the most durable marketing asset a small company can build.

Getting it read by the right people

• Email ten specific people before publishing — journalists, newsletter writers, association staff — with the headline finding in the subject line and the embargo date.
• Offer the chart as a file, not just a link. Editors use what is easy.
• Give trade associations the data cut for their members. They will circulate it because it is about their people.
• Post the five findings as text in the places your industry argues, and link to the method.
• Answer the criticism in public. The person who picks holes in your sample is doing you a favour, and responding well is more credible than the report itself.

Sources

• Google Search: creating helpful, reliable, people-first content
• What is GDPR? A plain summary
• KDP: Content guidelines, including AI-generated content
• Nielsen Norman Group: how users read on the web

Common questions