How to run shopping feeds audience research for multichannel shops

What to take away

Audience research now sits inside a compliance frame. The ICO's lawful basis guidance sets out the six lawful bases for processing customer data, so any survey, interview or feed segmentation plan needs a documented basis before you collect anything.
Research at the ASA and CAP shows regulators test advertising claims with real audience evidence, so the language in your feed titles and descriptions can be checked against how people actually read them.
The ID attribute in Google Merchant Centre remains the anchor of feed structure, which means audience segments are usually built on top of clean product identifiers rather than instead of them.
Most England-based feed teams should research in three passes: personas, then interviews, then surveys, and only then competitor checks.
Budget a fixed number of hours per pass. A small team can complete a first research cycle in roughly two to three weeks of part-time work.

Why audience research changed for feed teams

Feed work used to be a technical job. You mapped attributes, fixed disapprovals, and moved on. That has shifted. The people reading your feed are now the same people regulators and platforms ask about, and the evidence trail matters.

The ICO's guidance on lawful basis for processing personal data is the clearest example. It requires you to name a basis, such as legitimate interests or consent, before you handle customer records. Audience research touches those records constantly.

At the same time, feed data itself has become an audience signal. Platforms infer intent from titles, images and categories, so a misread description is a misread audience. Research tells you which words land.

There is a commercial reason too. Multichannel ecommerce teams that guess at audience needs tend to over-invest in channels that never convert and under-invest in the ones that do.

Decide what you actually need to learn

Before any survey goes out, write down the decision the research will change. If no decision changes, the research is a hobby.

Typical decisions in a feed context include which product categories to prioritise, which attributes to enrich first, which channels deserve more budget, and which audience segments justify their own feed.

Keep the list short. Three or four decisions is workable for a small team. Ten is not.

Each decision should map to a question you can answer with evidence rather than opinion. For example, if the decision is whether to split a feed by size, the question is whether size is a genuine filter shoppers use.

Turning decisions into research questions

Write each question so a stranger could answer it. "Do shoppers care about material?" is vague. "When choosing between two similar coats, which attributes do shoppers check first?" is answerable.

Then note the evidence type that would settle it. Some questions need interviews. Some need a survey. Some need platform data you already hold.

Choosing a method for each question

Interviews suit early, messy questions where you do not yet know the vocabulary. Surveys suit confirmation at scale. Competitor work suits category norms and pricing signals.

Do not run all three at once. Sequence them so each pass sharpens the next.

Start with buyer personas built on feed behaviour

A persona that only describes demographics is not useful for feed work. What matters is how the person shops, searches and compares.

Build personas around feed-visible behaviour. Does the person search by brand, by size, by price band, or by use case? Do they filter heavily or scroll? Do they arrive from a marketplace or a search engine?

Our guide to shopping feeds buyer personas in England walks through the structure in detail, including how to keep personas tied to attributes you can actually edit in a feed.

Aim for three to five personas. More than that and nobody remembers them, which defeats the purpose.

Give each persona a name, the decision it drives, and the feed attributes it depends on. That last column is what turns a persona into an action.

Run interviews before you run surveys

Interviews exist to find the words people use. Feed copy is written in words, so this pass pays for itself quickly.

Recruit six to ten participants for a first round. Mix recent buyers with people who browsed and left. The second group often teaches more.

Ask about the last purchase, not purchase in general. Specifics beat opinions. "Walk me through the last time you compared two similar products" gets usable material.

Record only what you need, and tell participants what happens to their data. The ICO's advice for small and medium organisations includes a self-assessment route that helps smaller teams check their handling before a project starts.

Our walkthrough of shopping feeds customer interviews in England covers recruitment, scripts and how to turn transcripts into feed edits.

After the round, list the exact phrases participants used. Those phrases belong in your titles and descriptions where they fit naturally.

Turn findings into a survey you can trust

A survey confirms what interviews suggested. It does not discover it. Running a survey first usually produces questions nobody cares about.

Keep it under ten minutes. Ask forced-choice questions where possible, because open text responses are hard to analyse at volume.

Pilot with five people before launch. If they hesitate on a question, rewrite it.

Sampling matters more than sample size for feed decisions. A hundred responses from your actual customer base beats a thousand from a general panel.

Our overview of shopping feeds survey tools: data and sources compares collection routes and the data sources worth combining with survey output.

Report percentages with the base number attached. "62% of 140 respondents" is honest. "62% of shoppers" is not.

Keep the data protection side clean

Audience research collects personal data the moment you can identify a respondent. That triggers obligations under the UK GDPR.

Name your lawful basis before collection begins. The ICO's lawful basis guidance for processing customer data sets out the options and the conditions attached to each, including the balancing test for legitimate interests.

Retention is the other common gap. Decide how long you keep recordings and raw responses, and write it down. A default of twelve months is workable for most projects, but the decision must be yours and it must be recorded.

If you plan to combine research data with feed performance data, say so in the privacy notice. Combining datasets changes the risk profile.

For smaller operations, the ICO's advice for small and medium organisations is a sensible starting point, because it frames data protection as a set of practical steps rather than a legal essay.

Check your feed structure before you segment

Audience segments only work if the underlying feed is consistent. Inconsistent identifiers produce segments that double-count or miss products entirely.

The ID attribute is the foundation. Google's specification for the ID attribute in product feeds explains that each item needs a unique, stable identifier, which is exactly what lets you join feed data to research findings later.

Audit before you segment. Check for duplicate IDs, missing identifiers and products that change ID when stock refreshes.

Once identifiers are stable, you can tag products by persona, by research theme, or by the attribute that interviews flagged as decisive.

Use competitor research to set the baseline

Competitor work tells you what the category treats as normal. That is useful context, not a strategy.

Look at how rivals structure titles, which attributes they surface early, and which categories they split into separate feeds. Note gaps rather than copying patterns.

Our guide to shopping feeds competitor research in England covers how to record findings so they stay comparable over time.

Keep the sample small and fixed. Ten competitors reviewed consistently beats fifty reviewed once.

Refresh the review quarterly. Feeds change, and a baseline from two years ago misleads more than it informs.

Test advertising claims against audience evidence

The words in a feed often reappear in ads. That means audience research can also support compliance work.

The ASA and CAP publish research on advertising effectiveness and compliance, which shows how audience evidence is used to judge whether claims are likely to mislead. If your feed says "best selling" or "number one", you need a basis for it.

Interview and survey output can supply that basis when it captures how shoppers interpret a claim. It is not a substitute for legal advice, but it is evidence.

Keep a simple record: claim, evidence source, date, and the person who approved it. That record is worth more than a folder of transcripts.

Where wider marketing research fits

The Data and Marketing Association publishes research on data and marketing practice that is useful for benchmarking how other teams handle audience segmentation and measurement.

Use it to sanity-check your approach rather than to set targets. Benchmarks from other sectors rarely transfer cleanly to feed work.

If your team also sells internationally, compare your England findings against market-level differences before assuming they carry. Our shopping feeds market guide for 2027 sets out the England picture and the areas where behaviour diverges.

Treat cross-market differences as a research question, not an assumption.

A research checklist for small feed teams

Work through this before each project. It is deliberately short enough to finish in one sitting.

Decision written down, with the person who owns it named.
Research questions phrased so a stranger could answer them.
Method chosen per question, with interviews before surveys.
Personas limited to five and tied to editable feed attributes.
Six to ten interview participants recruited, including non-buyers.
Survey piloted with five people and kept under ten minutes.
Lawful basis named and recorded before any collection.
Retention period set for recordings and raw responses.
Privacy notice updated if research data is combined with feed data.
Feed IDs audited for duplicates and stability.
Competitor sample fixed at ten and reviewed quarterly.
Claim evidence log started, with dates and approvers.

Print it, or keep it as a shared document. The value is in the repetition, not the format.

How long this takes and what it costs

A first cycle is not a big project. For a two-person team working part-time, allow two to three weeks.

Costs are mostly time. Participant incentives are the main cash item; for example, a team offering £25 per interview across eight interviews spends £200, which is an illustrative figure rather than a market rate.

Survey tools vary widely. A basic paid tier is often enough for a few hundred responses, so avoid annual commitments until you know the cadence you can sustain.

If you buy in research support, scope it by decision rather than by deliverable. "Help US decide whether to split the feed by size" is a better brief than "run some interviews".

Common questions

How many people do I need to interview?

Six to ten is enough for a first round in most feed projects. The goal is vocabulary and themes, not statistical confidence. Stop when new interviews repeat what you already heard.

Can I use existing customer data instead of running new research?

Sometimes. Order and search data can answer questions about behaviour, but it rarely explains why. Use existing data to frame interview questions rather than to replace them. Check your lawful basis covers the new use.

Do I need consent for every research activity?

No. Consent is one of several lawful bases. Legitimate interests can cover some research, provided you complete a balancing test and tell people what you are doing. Record the reasoning either way.

How often should I repeat the research?

Repeat interviews annually and refresh competitor checks quarterly. Surveys can run twice a year if you have a stable question set. Change the questions only when the underlying decision changes.