The four silos of market research — and why your view of the market is broken

This piece draws on a full Theia study of our own category — the market-intelligence market (UK + US), 71,000+ enriched snippets across web, YouTube and community sources, and a 468,801-row calibrated search-volume corpus across eight markets. The four-silo structure is the study's measured competitive landscape; every quotation is verbatim from its evidence base.

We ran a full market study on our own industry — the tools, platforms, and agencies a marketing leader uses to understand their market. Then we noticed something in our own data: a whole category of competitor never showed up.

The Amazon agencies — the firms that run brands' entire Amazon presence — were absent. Zero. Not because they don't matter, but because they live in a conversation that doesn't overlap with the one the survey platforms and SEO tools live in. Our study, seeded from the "market research" and "insights" vocabulary, simply never reached them.

That absence is the most honest finding in the whole study, because it is the problem: market research isn't one market. It's four silos that don't talk to each other. And if your view of your market is stitched together from all four by hand the night before a strategy meeting, your view is broken.

Market research agencies vs tools vs AI platforms — how should you choose?

Short answer: you shouldn't choose one of them, because the four silos don't combine into a market view. A survey platform, an SEO suite, an Amazon agency and an analyst subscription each answer a different question on a different dataset, and no one of them sees the whole customer. The reliable move is a single instrument that unifies demand, visibility, sales and perception on data you can verify — the room, not a fifth keyhole. The rest of this piece is the evidence for why the stack fails, and what replaces it.

The four silos

Each is excellent at its keyhole. None sees the room. A survey platform will tell you what a panel says it wants; it can't tell you what the same people search for, or what they buy on Amazon, or whether an AI assistant recommends you when they ask. Four vendors, four bills, four exports — and a human left to reconcile them into a story.

You don't have to take our word for the fragmentation — the AI assistants confirm it. Ask one today for the best UK market-research setup and it hands back a stack, one tool per silo:

"If your goal is specifically serving UK clients, the strongest overall stack in 2026 is generally considered to be: Brandwatch for social intelligence, GWI for audience insights, Attest for UK consumer surveys, quantilope for advanced quantitative research…" — surfaced verbatim from an AI Overview answer in the study

Five products to see one market. That is not a recommendation; it is a symptom.

That reconciliation is where strategy dies. By the time the four views are glued into one slide, they're stale, they don't agree, and nobody trusts the seams. The result is a market research function that is busy but not strategic — tactical answers to tactical questions, never a single coherent picture of the market.

The deeper problem: even inside a silo, the data isn't reliable

It's worse than fragmentation, because the inputs themselves are shaky. Take the one number everyone in the search silo quietly trusts — keyword search volume. Measured across our own 468,000-row demand corpus:

• 26% of singular/plural keyword pairs report byte-identical volume — the tools serve the whole "family" total to each variant, so summing a keyword list double-counts.
• Calibrating those figures against exact-term demand roughly halves the typical number.
• The language boundary leaks: English keywords score 96% of their UK volume in France — demand that isn't real.

And the newest silo — the AI-visibility tools promising to tell you whether ChatGPT recommends you — is often the worst offender. A practitioner in the study put it bluntly:

"Selling visibility scores averaged over invented prompt baskets, queried through API endpoints no consumer uses, weighted by modelled volumes nobody can verify, with woolly methods if they're published at all, is sketchy at best." — practitioner, surfaced verbatim from the study

So the silos don't just fail to combine. Each is reporting numbers that need correcting before they can be trusted — and some are reporting numbers that can't be checked at all. A disjointed market built on unverifiable data is not a foundation for strategy.

The alternative: one instrument, better data, organised around the customer

The fix isn't a fifth tool for the pile. It's a different shape entirely — a single instrument that concentrates the whole category's knowledge in one place. Three things have to be true of it:

1. It brings every signal together — demand, visibility, sales, and perception, from search and voice and Amazon and AI answers, in one view. The room, not a fifth keyhole.
2. It improves the reliability of the data, not just its quantity — de-merges the close variants, calibrates the volumes, corrects the language leak. A number you can decompose to its sources is a number you can defend to a board.
3. It organises around the customer, not the tool — the journey (awareness → demand → conversion → retention) is the spine, so the question is always "what does the customer do," never "what does this particular instrument measure."

This is what an AI-native market intelligence platform is for, and it's why the category is being rebuilt rather than incrementally improved. The winner won't be the best survey tool or the best keyword tool. It'll be the one that made the four silos into one picture and made the picture trustworthy.

What you actually get from Theia

"One instrument" is abstract until you see what it delivers. A Theia engagement gives a consumer brand:

• The whole market in one map. A full category study: who competes, on which keywords, what customers search, say, and buy, and which sources AI cites — demand, visibility, sales, and perception in a single picture, across markets and languages.
• Calibrated demand you can defend. Search volumes corrected for the close- variant and language distortions above, and decomposable back to their Google Ads, Trends, and Search Console sources. A number you can put in front of a board and stand behind.
• Competitive positioning. Exactly where you win and lose against each rival, on the features and use-cases customers actually care about — not a quadrant, a read on the real contest.
• Content that ships, not just insight. Amazon listing copy generated and optimised across seven languages, grounded in real consumer perception; web pages written in-brand and evidence-checked. Insight that becomes execution, at scale, without a hallucinated claim.
• Share of search and whitespace. The metrics behind where to invest: what you own, what a rival owns, and what nobody owns yet.
• AI-answer visibility (GEO). Whether the AI assistants recommend you, tracked over time — measured on the same instrument we used to measure ourselves.
• A chatbot with every study. Ask it unlimited questions, answered only from the study's evidence. The deck answers today's questions; the chatbot answers next month's.

And the one thing none of the four silos can claim: every number traces to its evidence. Grounded, decomposable, auditable — because a market view you can't defend isn't a strategy, it's a guess.

The tell

If you want to know whether your market view is whole or siloed, ask a simple question: when a customer researches your category, do you know what they search, what they say, what they buy, and what the AI recommends — from one source, today? If the answer takes four logins and a spreadsheet, you're not seeing your market. You're seeing four slices of it and guessing at the middle.

We found that out about our own market by accident, when a competitor silo failed to appear. It's a good thing to learn about a market. It's a dangerous thing not to know about yours.

You can interrogate a live Theia study yourself: the weight-loss market chatbot answers only from its measured evidence, with quotes — the "one instrument" idea, running. And if you want your own category read this way, that's what we do.

Related reading

• The State of AI Market Research — the full study this piece draws on (the pillar).
• How AI answers decide their shortlist for market research — GEO, the newest silo, measured.
• Synthetic respondents vs structured evidence — why the voice silo's newest promise needs grounding.
• Ask the market anything — the chatbot that ships with every study.

References

1. Theia, State of AI Market Research study, 2026 — UK + US, 71,000+ enriched snippets (web, YouTube, community) + a 468,801-row calibrated search-volume corpus across eight markets. The four-silo structure is the study's measured competitive landscape; quotations are verbatim from its evidence base.
2. Google Ads Help — About close variants, on the aggregation of singular/plural and near-variant keywords in Keyword Planner volume.
3. Aggarwal et al., GEO: Generative Engine Optimization, KDD 2024 — on measuring and improving how generative engines cite sources (the method behind this page's structure).

---

Method: findings drawn from a full Theia study of the market-intelligence category (UK + US), 71,000+ enriched snippets across web, YouTube, and community sources, and a 468,801-row calibrated search-volume corpus across eight markets. The four-silo structure is the study's measured competitive landscape; the Amazon-agency silo's absence from the initial collection is reported as-is.