Use AI for market research to organize a supplied evidence packet and expose questions it cannot answer. Start with the research decision, keep providers separate from individual offers, and make every proposed conclusion show its evidence scope.
This guide uses a fictional practice packet. Provider P1 owns two notices: N1, “Intro formulas, 60 minutes,” and N2, “Intro charts, 90 minutes.” They are authored examples, not real market observations. The finished deliverable is a question-to-method protocol, an offer ledger, and a human-reviewed claim brief. No actual AI run or customer interview has taken place.
Define the decision before requesting a summary
The practice question is: Is there unmet demand for a weekend beginner spreadsheet workshop? That question is not answered by finding a few classes. Before real research, specify the proposed service, intended participants, geography or delivery reach, and observation window.
| Scope field | Actual definition |
|---|---|
| Proposed service and beginner level | [TO CONFIRM] |
| Intended participant segment | [TO CONFIRM] |
| Geography or delivery reach | [TO CONFIRM] |
| Observation window and as-of date | [TO CONFIRM] |
| Decision the research should inform | [TO CONFIRM] |
These definitions belong in the native protocol. “Local” is not a research boundary until you say which locality and why it matters. “Beginner” needs a service definition rather than an assumption that all spreadsheet users have the same needs.
The SBA distinguishes research using existing sources from research conducted directly with consumers and asks businesses to consider demand, location, alternatives, and market segment. That distinction supports our choice of methods; it does not verify the fictional notices or establish demand. SBA market research guidance.
Record offers without multiplying providers
| Notice ID | Provider ID | Supplied topic | Supplied duration | Date / location / price / enrollment / actual URL |
|---|---|---|---|---|
| N1 | P1 | Intro formulas | 60 minutes | Unknown |
| N2 | P1 | Intro charts | 90 minutes | Unknown |
There are two supplied session notices and one supplied provider. Counting two rows as two companies changes the unit of analysis. Nor does a single provider in a small packet establish that the market has only one competitor.
The two durations are offer attributes. They do not show attendance, preference, willingness to pay, satisfaction, or how long a learner actually needs. A notice can tell you what is offered without telling you whether enough people want a different offer.
When replacing this fixture with real sources, open each permitted source and verify its provider identity, wording, publication or observation date, and relevance to the chosen scope. Keep a source URL and source ID with the recorded fact. If price is absent, record “not supplied”; do not invent a free price or infer it from another session.
Map each question to a method
| Question | Current evidence | What it can support | Next method |
|---|---|---|---|
| Q1: Which topics and durations are listed? | N1 and N2 | Formula introduction, 60 minutes; chart introduction, 90 minutes, within this fictional packet | Verify actual notices and record their scope/date |
| Q2: Is there unmet weekend beginner demand? | No demand evidence supplied | Unknown | Direct research into participants' recent needs and alternatives |
| Q3: What other alternatives serve the target segment? | Only P1 supplied | No complete competitive inventory | Bounded secondary search using defined segment, reach and window |
The absence of a weekend notice in N1 and N2 is an absence in the supplied packet. It is not proof that no weekend class exists anywhere. Even a carefully bounded search would need its coverage and limits reported.
Do not turn the unknown demand field into a zero-demand estimate. “No evidence supplied” and “evidence of no demand” are different conclusions.
Audit a plausible-looking draft
The following claims are intentionally authored to resemble mistakes in an AI draft. They are not output from a model call.
| Draft claim ID | Original draft claim | Evidence review | Human disposition |
|---|---|---|---|
| C1 | “200 local customers want formulas.” | No people, locality, preference or count evidence | Unsupported; remove from conclusions, retain in review history |
| C2 | “There are two competitors.” | Both notices belong to P1 | Conflicts with provider count in packet; revise to two notices from one provider |
| C3 | “No weekend classes exist.” | Neither notice supplies schedule coverage | Overgeneralization; report weekend coverage as unknown |
Keep the original claim beside the disposition. A clean summary alone hides what changed and why. A reviewer should be able to trace the correction to the ledger.
A restrained brief would say: “The supplied fictional packet lists two sessions from P1: formulas for 60 minutes and charts for 90 minutes. It contains no demand or weekend-coverage evidence. Further source verification and direct research are needed before answering the proposed weekend-workshop question.”
That brief is useful because it names the next uncertainty. It does not award an opportunity score or recommend a price based on invented customers.
Prepare a direct question that does not supply its own answer
For a later interview, start with: “Tell me about the last spreadsheet task you wanted help with.”
Follow the actual account: what the task involved, what the person tried, where they sought help, and what happened. Timing preferences can be explored after a real need has been described. Avoid beginning with “Would you attend our great weekend workshop?” and treating a polite hypothetical response as measured demand.
Keep the actual recruitment, participant permission, answers, and analysis blank:
| Direct-research field | Actual record |
|---|---|
| Target participant and recruitment approach | |
| Permission/participation record | |
| Recent-task account | |
| Existing alternatives used | |
| Timing needs, if discussed | |
| Researcher's interpretation and limits |
These questions are an original proposed protocol, not a completed sample or a universal interview standard. Do not generate imaginary respondents to fill the empty table. A small convenience group would also need its limitations reported before any wider inference.
Use a supplied-source prompt with review gates
Give the assistant the scope record, verified source packet, and unit definitions. Ask it to attach each proposed conclusion to a source or mark it unknown. Human review still needs to open the actual sources and examine whether the quoted fact supports the claim.
The prompt below uses only the fictional packet. It is complete enough to prepare the brief while refusing unsupported demand conclusions. It is not executed here.
Prepare an English market-research review brief from supplied evidence
only. Do not browse for this exercise, call tools or services, invent
market data, generate respondents, conduct interviews, or execute a
separate model workflow.
Practice decision:
Is there unmet demand for a weekend beginner spreadsheet workshop?
Actual service/level, target segment, geography/delivery reach,
observation window and decision boundary are TO CONFIRM.
Fictional practice evidence:
Provider P1 owns both notices.
N1: "Intro formulas, 60 minutes."
N2: "Intro charts, 90 minutes."
Notice dates, locations, prices, enrollment, schedules and actual URLs
are unknown. These authored fixtures are not real market observations.
Units:
A notice/session offer is not a distinct provider.
The packet has two notices from one supplied provider.
It is not a complete market inventory.
Offer topics/durations are not customer preferences or demand counts.
Questions:
Q1: Which listed topics and durations does the packet support?
Q2: Is there unmet weekend beginner demand?
Q3: What other alternatives serve the eventual target segment?
Retain these intentionally authored AI-like draft claims for review:
C1 "200 local customers want formulas."
C2 "There are two competitors."
C3 "No weekend classes exist."
These are not observed model output.
Deliver:
1. Native scope-definition record with unresolved fields.
2. Offer ledger keeping notice ID, provider ID, topic, duration, source,
date, location, price, enrollment and coverage separate.
3. Question-to-evidence/method map: secondary source review for offers
and alternatives, direct research for actual participant needs.
4. Claim review preserving C1-C3 alongside evidence, scope, unknowns
and human disposition. C1 unsupported; C2 conflicts with shared P1;
C3 overgeneralizes missing schedule information.
5. A short corrected brief limited to the fictional packet.
6. A proposed nonleading interview opening:
"Tell me about the last spreadsheet task you wanted help with."
Keep actual recruitment, permission, answers and interpretations blank.
Do not invent 200 customers, demand, market size, revenue, attendance,
price recommendations, representative sampling, or an opportunity score.
Missing weekend notices do not prove market-wide absence.
Missing information is unknown, not zero or a negative finding.
For actual later research, require a human to open permitted sources,
verify provider identity, wording, scope/date and URLs, and retire old
conclusions when evidence or scope changes.
Mark actual research, model outputs, interviews and controls pending.
No AI accuracy, automated verification or success guarantee.
End with a separate NOS use area for saving/rendering a prepared
fictional or sanitized HTML brief. Native sources/protocol and human
review remain authoritative; rendering does not collect market evidence.
Use NOS for the prepared evidence brief
NOS can save and render prepared HTML. A fictional review copy can put N1/N2, the provider count, and C1–C3 beside their dispositions so a reader sees exactly what the packet supports.
Keep the native source ledger, research protocol, and human verification record authoritative. Actual research sources, interviews and AI output remain to be collected; the local controls review only the supplied fictional packet.
Explore NOS with the prepared HTML evidence review. Leave the team with a supported offer comparison and a specific next research question, rather than an unsupported demand estimate.