Market Intelligence
Why this exists
Most AI "market research" is one prompt and a confident answer. You can't tell which figures came from a real source, which the model remembered from its training data, and which it invented. A market size with no publisher and no date isn't something you can act on.
Market Intelligence works the other way round. It collects evidence first: web pages, live search data, crawls of competitor sites and your own store's figures. It turns that evidence into individual claims, checks each claim against its sources, and then writes the report only from claims that passed. If a paragraph can't point to a checked claim, it's removed.
The result is shorter than a one-prompt essay, and you can check every sentence in it.
Where to find it
Open Market Intelligence in the app navigation. The hub has four tabs:
| Tab | What it does | Plan |
|---|---|---|
| Market | Research your product category in the markets you sell to | Pro+ |
| Competitors | X-Ray each competitor in your competitor list | Pro+ |
| SWOT | Strengths, weaknesses, opportunities and threats, with evidence for each | Starter+ |
| Research runs | Every run, how it ended, and (with permission) what it cost | Starter+ |
The top of the page shows how many of each run type you've used this month, how many runs are in progress, and (if you can see spend) your research spend against your plan's allowance.
How a run works
- Admission. Before anything is spent, the app checks your plan, this month's allowance, your research budget and how many runs you already have in progress. If any check fails you get a plain message saying why. Nothing is charged.
- Collect. Measured evidence that needs no web research: your store's figures, competitor price and crawl data you already track, and live search data where available.
- Research. For market reports and X-Rays, a research model searches the web and reads pages, one question at a time. Only passages that cite a source are kept.
- Extract and verify. Findings become individual claims. Each claim is checked for independent corroboration, disagreement between sources, and age.
- Write and publish. The report is written from verified claims. Any paragraph without a verified claim is removed, and the result is saved as a new version.
A SWOT skips the web research. It works from your store's data plus any market reports and X-Rays you already have, so it's quick and cheap.
Run statuses
| Status | Meaning |
|---|---|
| Queued | Admitted and waiting for a research worker |
| Running | In progress. The progress bar moves as each question is answered |
| Done | Published in full |
| Partial | Published, but part of it couldn't be completed. The report says which part and why |
| Failed | Nothing was published. The Research runs tab shows the reason |
| Cancelled | You stopped it. Anything already spent is recorded |
Cancelling. A queued run is cancelled straight away. A running run stops at its next checkpoint, between research questions or stages, so you're charged only for what it already did.
Asking twice. If you request exactly the same report twice on the same day (UTC), you get the run that's already there rather than paying for a second one. A run that failed or was cancelled doesn't block a retry.
How to read a report
The four labels
Every figure and every paragraph carries one of these labels:
| Label | Meaning |
|---|---|
| Measured | From your store, a crawl, or a provider that measures (for example live Google results or Google Ads search volume) |
| Reported | A third party stated it. The source page and its date are cited |
| Estimated | Inferred or modelled, not measured. Always labelled, never shown as fact |
| Unavailable | We couldn't collect it. The reason is shown, for example "the platform's DataForSEO account is out of credit" |
An unavailable figure is never quietly replaced with an estimate. A gap that tells you why it's a gap is more useful than a number you can't trust.
How well supported a claim is
Open Show evidence under any paragraph to see the claims it rests on, the quoted passage from each source, and a link to the source page.
| Badge | Meaning |
|---|---|
| Corroborated | At least two independent publishers support it. Two subdomains of one publisher count once |
| Single source | One publisher supports it |
| Sources disagree | Sources give materially different figures (more than 25% apart) for the same thing in the same period. Both are shown; the app doesn't pick one |
| Stale | The figure describes, or was published, more than 24 months ago |
| Low confidence | Weak or disputed evidence, or a figure that doesn't appear in the passage it cites |
Confidence is worked out from the evidence, not from what the model says about itself. A model can lower its own confidence, but it can't raise it.
What the header tells you
The report header shows how many claims the report rests on, how many are corroborated, single-source or contested, and how many sources were used. If the run was cut short, a banner lists Parts of this report could not be completed with the reasons. If this is a newer version, a second banner summarises what changed since the previous one.
At the foot of the report, any section with no strong enough evidence is named: "No cited evidence strong enough to report on: …". That's the engine telling you it found nothing it could stand behind. It isn't an error.
Versions and exports
Every run creates a new version. The previous version is marked superseded but stays readable. A newer version summarises what changed: figures that moved, new findings, and findings that are no longer supported.
Each report has two exports:
- Report (.md): the report as written, with every citation as a numbered footnote and a source list with URLs and dates. The evidence goes with the report when you share it.
- Evidence (.csv): one row per claim, with its support, label, confidence and sources. Useful for checking a report line by line.
Plan tiers
| Free | Starter | Pro | Agency | Enterprise | |
|---|---|---|---|---|---|
| SWOT runs per month | — | 4 | 8 | 30 | 100 |
| Market reports per month | — | — | 4 | 20 | 100 |
| Deepest research allowed | — | — | Standard | Deep | Deep |
| X-Ray runs per month | — | — | 10 | 50 | 200 |
| Competitors you can X-Ray | — | — | 5 | 25 | 100 |
| Monthly research budget | — | $2 | $30 | $150 | $1,000 |
| Scheduled monthly refresh | — | — | — | ✓ | ✓ |
Allowances and the budget reset on the 1st of each month (UTC).
Why SWOT starts at Starter. It runs on data your store already holds and makes no web searches, so it costs cents. Market reports and X-Rays search and read the web, which is where the cost is.
The research budget
Research is paid per use: model tokens, web searches, and billed data such as DataForSEO search volumes. When you start a run:
- The app estimates the cost from the report type and depth.
- The estimate is reserved against your monthly budget.
- The run is stopped if it reaches 1.5 × its estimate. Whatever it gathered so far is published as a partial report.
- When the run ends, the reservation is replaced by the actual cost.
A run that fails before spending anything doesn't use up any of your monthly allowance.
Research also counts against your store's general AI budget. If that runs out mid-run, research stops and the report is published as partial with that reason.
Scheduled refresh
On Agency and Enterprise, any market report, X-Ray or SWOT you've run at least once can be refreshed monthly. Use Refresh monthly on the report. A refresh asks the same question as the run it was created from: same category, markets, depth and focus questions.
A scheduled refresh is just a request made on your behalf. It goes through the same plan, allowance and budget checks as a click. If a check refuses it, the schedule shows Last refresh not run with the reason:
| Refusal | What the schedule does |
|---|---|
| Monthly allowance or budget used up | Waits until the 1st of next month |
| Your plan no longer includes it | Switches itself off |
| A platform problem (research switched off, queue unavailable) | Tries again in about 6 hours |
Who can do what
| Permission | Allows | Default roles |
|---|---|---|
intel:report:read | Open the hub and read reports | Owner, Admin, Marketer, Analyst |
intel:research:run | Start and cancel runs, turn scheduled refresh on or off | Owner, Admin, Marketer |
intel:report:export | Download the Markdown and CSV exports | Owner, Admin, Marketer, Analyst |
intel:swot:manage | Accept or dismiss SWOT items, propose and decide actions | Owner, Admin, Marketer |
intel:spend:view | See research cost, estimates and the monthly budget | Owner, Admin, Marketer, Analyst |
Starting a run spends money, so it's a separate permission from reading. Analysts can read, export and see spend, but can't start runs.
Messages you might see
| Message | What it means | What to do |
|---|---|---|
| Market Intelligence is temporarily switched off | The platform has paused new runs. Existing reports stay readable | Wait, or contact support |
| Web research is not configured | The platform's research provider isn't set up. SWOT still works | Contact support |
| You have used all N … runs included this month | Monthly allowance reached | Wait for the 1st (UTC) or upgrade |
| This run would exceed this month's … research budget | The estimate doesn't fit in what's left of your budget | Try a shallower depth, wait for the 1st, or upgrade |
| You already have 3 research runs in progress | A store can run 3 at once | Wait for one to finish |
| An identical report was already requested today | Same request, same day | Nothing. You're shown the existing run |
Best practices
- Fill in your positioning canvas first. It pre-fills the market category and gives the research better context.
- Start with a SWOT. It's the cheapest run and shows you which store signals have enough data. Then run a market report and a couple of X-Rays and rebuild the SWOT. Its opportunities and threats widen automatically.
- Use Standard depth for your first market report. Go deeper only when the report shows sections with no strong enough evidence that matter to you.
- Treat "Sources disagree" as a finding. Market-size figures from different publishers routinely differ by a lot. The range is more honest than one number.
- Read the evidence before acting on a figure. Every claim is one click from its source, and that click is the point of the product.
See also
- Market reports: researching your category
- Competitor X-Ray: profiling a competitor across 11 dimensions
- Evidence-based SWOT: SWOT, TOWS strategies and action proposals
- Competitor intelligence: the competitor list the X-Ray profiles
- Positioning: the canvas that pre-fills your category