A useful AI market research report answers a business decision and shows the evidence behind its recommendation. It defines the audience and time period, separates observations from assumptions and makes missing information visible.
Start with a question you can act on. “Research AI marketing” is too broad. “Which two tools should a small ecommerce team trial for producing a campaign from one brief?” gives the work a boundary and an outcome.
Use the template below or download the research report template. The worked example is a desk-research illustration, not customer research or a hands-on product benchmark.
Define the decision before collecting sources
Write the decision, audience, geography and period at the top of the brief. Include what is outside scope.
For a tool-selection question, a useful boundary might be:
We are a small ecommerce marketing team choosing a tool to trial for a launch campaign. We need copy, product-image concepts and a short narration. We are comparing current public product descriptions, not measuring creative quality, ad performance or customer demand.
That last sentence protects the recommendation from becoming broader than the evidence. Public documentation can describe a feature. It does not tell you how your team will perform with it, what buyers think or whether a campaign will sell.
Then name the evidence required. A pricing decision needs current terms and the actual workload. A positioning decision needs customer language and alternatives. A launch decision may need several kinds of evidence, including direct conversations with prospective buyers.
Copy the report template
TITLE AND DATE
Research question:
Decision owner:
Geography, audience and period:
Sources checked through:
DECISION SUMMARY
Recommended next action:
What supports it:
What could change it:
SCOPE AND METHOD
Included questions:
Excluded questions:
Source types and selection rules:
Known access limitations:
FINDINGS
For each finding:
- observation;
- supporting source and exact relevant passage;
- source date and access date;
- qualification or conflicting evidence;
- implication for the decision.
COMPETITOR OR OPTION TABLE
Option, relevant capability, terms to check, evidence and unknowns.
ASSUMPTIONS AND GAPS
Unverified assumptions:
Questions requiring customer interviews or an actual test:
Information not found:
RECOMMENDED NEXT STEP
Task, owner, time limit, budget and acceptance criteria.
SOURCE REGISTER
Source ID, publisher, title, URL, dates and supported claims.
Keep evidence close to the finding. A source list at the end is helpful, but a reader should not have to guess which link supports a particular number or conclusion.
Choose sources by the question they can answer
Prefer primary material for product availability, terms, official figures and policy. Use interviews for customer experience. Use secondary analysis to find context and alternative explanations, then trace consequential claims to their origin when possible.
Ask what each source cannot establish. A vendor case study may describe one customer's experience; it does not establish a typical result. A list of tools proves that someone listed them, not that the list reflects measured market share.
If several articles repeat the same statistic, look for a shared source. Ten pages can still represent one unverified claim. Record the original source rather than treating the repetitions as independent confirmation.
For a time-sensitive report, distinguish the date an event occurred, the date an article was published and the date you accessed it. “Updated recently” is not enough if the underlying data is old.
A worked report for a campaign-tool decision
Question: What should a small ecommerce team test before moving its launch-asset work into a broader AI workspace?
Audience: A hypothetical team that needs campaign copy, product-image concepts and a short voiceover.
Method: Review first-party public descriptions available on 7 November 2026. Identify relevant capabilities and design a common trial. Do not infer quality, savings or conversion performance from feature availability.
Evidence from public descriptions
| Option | Relevant documented description | Question for a trial |
|---|---|---|
| AI Magicx Campaign kit | Describes copy, channel posts, product images and voiceover from a brief | Does the resulting package preserve our facts and product details? |
| SuperMIA | Describes turning a brief into multi-channel campaign material | Which of our required formats can the selected plan actually deliver? |
| Jasper | Describes brand and company context supporting marketing work | Can our approved context be applied consistently to the task we need? |
Sources: AI Magicx Campaign kit, SuperMIA's campaign page and Jasper's product overview. These pages describe the products' positioning and capabilities; they do not constitute a comparative output test.
Findings
Finding 1: A brief-to-campaign workflow is relevant to the decision. The public descriptions support considering tools that organize more than a single writing task. The team should test the complete package rather than choose solely by chat-model availability.
Finding 2: “Campaign” can mean different outputs. A product may focus on written material, visual production, distribution or brand context. The team's trial needs an explicit deliverable list so that it compares equivalent work.
Finding 3: Product accuracy remains an acceptance criterion. The team's own listing requires accurate product details. A plausible generated visual is not enough to approve it. The trial should compare every candidate with a real reference photograph.
The third finding is a requirement of this hypothetical buyer, not a performance claim about any vendor.
Recommendation
Run one bounded trial using an owned product, approved facts and a common brief. Choose two relevant candidates after checking current access and terms. Ask each for the same written deliverables and supported media, record manual handoffs and review the actual costs.
Approve a broader migration only if the team can produce usable work, preserve its sources, understand its spending and complete its review process. If the needed visual or distribution stage remains in a specialist tool, include that handoff in the decision.
Unknowns
This desk review does not establish output quality, actual task cost, team adoption, customer preferences or campaign results. It supports a trial design. The next decision requires observed work.
This is a complete and useful research conclusion: it narrows the options and defines the test without pretending that documentation answered every question.
Build a claim-to-source table
For each important claim, keep a record like this:
| Field | What to record |
|---|---|
| Claim | The exact statement you plan to make |
| Source | Publisher, title and direct URL |
| Supporting passage | The words, table or result that support it |
| Dates | Publication or data date, plus access date |
| Scope | Geography, sample, product version or plan |
| Status | Supported, qualified, conflicting or not found |
If the source supports only part of a sentence, narrow the sentence or add another source. A link mentioning a market does not support a market-size estimate. A plan description does not support a claim that customers achieve a particular result.
Keep “not found” as a legitimate status. It prevents an assistant from filling an empty cell with a likely-sounding answer.
Ask an AI assistant to research within the boundary
Prepare a desk-research report for this decision:
[decision and audience]
Scope:
[geography, period, included questions and exclusions]
Use primary sources for product facts and current terms.
For each finding, include the direct source, relevant passage,
source date if available and access date.
Separate observations, interpretations and assumptions.
Do not invent market size, search volume or customer behavior.
Mark unsupported fields as "not found".
Identify questions that require interviews or a practical test.
End with one bounded next action and what would change it.
If you are using AI Magicx Auto and research, check the plan requirements and the estimate before starting paid research. A longer report or more links does not automatically make the conclusion better. The useful measure is whether the sources support the decision.
Check the report before sharing it
Open the sources for consequential findings. Read enough surrounding material to check conditions, dates and context. Verify that a quotation is real and that a number has the right unit and denominator.
Then test the recommendation against the gaps. If the missing evidence could reverse the decision, the recommendation should be conditional or should call for a smaller next step.
Use these questions:
- Does the report answer the stated decision?
- Are observations distinguishable from interpretations?
- Do the cited passages support the actual wording?
- Are conflicting sources explained rather than hidden?
- Are estimates, examples and assumptions labelled?
- Is the next action proportionate to the evidence?
For source documents, the PDF question-answering guide shows how to ask for bounded answers and inspect the underlying text.
Know what requires a customer conversation
Public desk research can help you identify competitors, compare published offers and prepare questions. It cannot directly tell you why your intended customers choose a product, which tradeoffs they accept or what they would pay for your proposal.
Turn those gaps into interview questions. Instead of asking “Would you use an AI campaign tool?”, ask about a recent campaign: how it was made, where work stalled, which tools were involved and what the buyer considered changing.
Keep the interview findings separate from vendor descriptions. Both can matter, but they answer different questions.
Keep the research useful over time
Give the report an owner and a review trigger. Pricing, product availability and competitor offers may need checking before the next purchase decision. A historical market observation may remain useful if its original date stays visible.
For recurring updates, use the weekly competitor brief workflow. Begin with an approved source list and compare against a saved prior report rather than relying on an assistant's recollection.
Start with the report template, define one decision and build the source table as you work. Then review the research workflow in Auto if you want help collecting and organizing the evidence.