Source-linked research workflow

Use an AI market research agent to make one founder decision clearer.

Soloop's Analyst runs approved market and competitor research, separates observations from inference, and returns evidence to the same project context. It is designed for startup decisions—not synthetic interviews or automatic proof of demand.

Start with a project See the workflow

Qualification boundary

Give the research agent a decision, not a topic cloud.

The same market can support hundreds of facts. A decision boundary tells the agent which evidence is material.

Use this workflow when

You need to choose between two customer segments, claims, channels, or product moves.
You want current sources and explicit unknowns rather than a confident unsourced summary.
You have an artifact the research can evaluate in market context.
You will use research to design a real test rather than declare the idea validated.

Do not count this as proof

You need statistically representative survey data or synthetic participants presented as real customers.
You want an exhaustive market report without a near-term founder decision.
You expect competitor pricing or availability to remain accurate without a dated refresh.
You want the agent to publish, email, spend, or change code without the applicable approval path.

Smallest complete workflow

Run a source → conclusion → decision handoff.

The research module stays replaceable because the output contract is stable: decision, evidence, inference, unknowns, and next test.

01
Frame the founder decision
State the choice, the current artifact, the audience, and what would change under each plausible answer.
02
Approve the research scope
Review the sources, recency, geography, competitor depth, budget or paid-service boundary, and completion criteria.
03
Collect primary evidence first
Prefer official product pages, documentation, pricing, terms, public records, and directly attributable statements. Use discussions to find questions and language, not as universal facts.
04
Label every evidence level
Keep company claims, observable facts, user reports, Analyst inference, and missing evidence distinct.
05
Test the conclusion against counterevidence
Search for facts that would reverse the recommendation. Mark unavailable evidence as not established rather than filling the gap.
06
Return one decision handoff
Summarize what changed, confidence and limitations, the strongest unknown, and one approved action capable of producing a higher-grade signal.

Inspectable control

Example approval card — one-competitor deep research

This is an illustrative task card, not a claimed customer result or completed experiment.

Objective: Decide whether an AI app builder is a direct demand competitor or a build-stage substitute for a founder operating workspace.
Action: Review one competitor's official homepage, documentation, pricing destination, and product boundaries; return a dated evidence ledger and decision implications.
Target: One named competitor and its approved official domains. No authenticated access, purchase, trial, or contact with the company.
Acceptance: Every material statement links to a source; claims, observations, inference, contradictions, and unknowns are separate; volatile facts carry an evidence date.
Stop condition: Stop before login, payment, account creation, or any source outside the approved public scope. Report unavailable evidence as not established.

ReviseRejectApprove exact scope

Analyst-style decision handoff

Return evidence in a stable, reviewable shape.

The fields below organize the dated official-source demonstration on this page. They model the expected output contract; they are not a claim that a customer task ran.

Decision: Is an AI app builder a direct demand competitor or primarily a build-stage substitute for a founder operating workspace?
Observed: Lovable documents natural-language full-stack building and deployment; Replit documents planning without editing code or data.
Inference: Both can absorb build and planning work, but the cited sources do not establish Soloop's complete proposal → approval → execution → signal loop.
Counterevidence: Planning inside a builder is already established, so Soloop cannot claim that planning alone is differentiated.
Unknown: Whether either workflow consistently connects distribution experiments and returned demand signals to the next founder decision is not established by these sources.
Next decision: Test whether founders value a smaller approval-and-signal loop after they already have a product, rather than competing on broader building capability.

Decision scorecard

Judge the report by decision quality.

A long report is not automatically a useful report. These checks keep the output reviewable.

Coverage

The sources address the customer, job, proof, exchange, and friction relevant to the decision.

Traceability

A reviewer can open the source behind every material observation.

Separation

Claims, observations, user reports, inference, and unknowns remain distinct.

Counterevidence

The report records facts that weaken or reverse the preferred conclusion.

Decision change

The evidence changes a customer, claim, priority, test, or stop decision.

Next signal

The handoff proposes a bounded action that can produce stronger behavioral evidence.

Research proof format

Example source-linked evidence excerpt

This demonstration uses real official sources checked in Soloop's September 1 competitor research. It shows the expected evidence structure; it is not a customer result or proof of demand.

Lovable describes a full-stack AI development platform that builds, iterates, and deploys web applications through natural language and real code.

Treat Lovable as a strong build-stage substitute; do not infer that it provides Soloop's complete approval-and-signal operating loop.

Lovable documentation

Replit documents Plan Mode as a way to brainstorm and plan with Agent without editing code or data.

Planning inside a builder is established. Whether planning stays connected to distribution experiments and returned company signals requires separate evidence.

Replit Starter plan documentation

Soloop's public manual describes approved research, X/Twitter, beta cold email, manual Reddit, plan-only ads, and sandboxed coding as distinct execution boundaries.

Compare systems by the complete proposal, approval, execution, evidence, and next-decision loop—not by the number of agent labels.

Soloop product manual

Current product boundary

Plan with what actually runs today.

Secondary research can establish public facts and support a hypothesis. It cannot prove that reachable customers experience the problem, will adopt the workflow, return, or pay. Those questions require direct behavioral tests.

Research

Live

The Analyst can execute approved market, competitor, and prospect research and return durable results and signals.

X / Twitter

Live with OAuth

A later approved Social task can publish through a connected account and collect replies.

Cold email

Beta · Gmail required

Research may support targeting, but email execution requires Gmail and its own approval path.

Manual publication

Public Reddit action remains with the founder even when research or a draft is prepared.

Plan-only

Research can inform an ad hypothesis; Soloop does not buy or manage ads today.

Coding

Live after approval

A research conclusion can hand off an approved product change to the sandboxed Coding Agent.

Continue the loop

Market research for startupsUse the evidence ladder and one-page decision output.SaaS competitor analysisApply the six-column founder matrix to one close competitor.Validate a SaaS ideaTurn the research conclusion into one falsifiable test.

FAQ

Questions before the next move.

What does an AI market research agent do?

It gathers and organizes evidence for a defined decision, cites the source behind material findings, labels inference and unknowns, and returns a conclusion plus the next test.

Can AI market research replace customer interviews?

No. It can improve the question, target, context, and hypothesis. Interviews and behavioral tests are still needed to learn whether reachable customers experience the problem and will change behavior.

Does Soloop run research without approval?

Consequential and paid or live work follows the applicable approval path. The task should show its objective, scope, sources, acceptance criteria, and stop condition before execution.

Bring the product you already have

Choose one move. Review it. Learn from the result.

Start with a URL, repo, landing page, demo, deck, file, or rough product.

Start with a project