Skip to content

AI Fundamentals for Marketers & Creators · Hands-on marketing workflows · lesson 10 of 16 · 13 min

Market and audience research with AI

Research is where AI saves marketers the most time

Before AI, a decent competitor scan or audience study took days. Now a deep research run, a review-mining session and a persona draft can be done in an afternoon. The catch: AI research is only as good as your brief, your sources and your checks.

Three research jobs, three tools

| Job | Best tool | Output | |---|---|---| | Market and competitor scan | Deep research (ChatGPT, Claude Research, Gemini Deep Research, Copilot Researcher) or Perplexity | Cited report: positioning, offers, price points, channels | | Voice-of-customer mining | File analysis on your own reviews, survey answers, support tickets, social comments (anonymized) | Themes, pains, desires, exact customer phrases | | Personas and hypotheses | A general assistant using the two outputs above | Draft personas labeled as hypotheses to validate |

Step 1: brief the market scan

Research brief
Decision: [e.g. how to position our new meal-prep service in Riyadh]
Questions: who are the 5 main competitors; how do they position (promise, audience, price range,
channels, offers); what do customers praise and complain about publicly?
Scope: Riyadh and Jeddah, sources from the last 18 months.
Sources: company websites, app store listings, reputable regional media, official statistics.
Treat anonymous forums and vendor "top 10" blogs as low confidence.
Output: competitor table (name, promise, audience, price range, channels, source link);
5 insights; gaps in the market; what you couldn't verify.

Read the research plan if the tool shows one. Afterward, spot-check five links.

Step 2: mine the voice of the customer

Your own customers' words are the most valuable research you have, and nobody else has them.

Attached: [N] anonymized reviews / survey answers / support messages.
1. Group into the top 6 themes (praise and complaint separately), with rough counts.
2. For each theme, give 3 exact customer phrases (verbatim quotes).
3. List the words customers use for the problem we solve (their language, not ours).
4. Note anything surprising or contradictory.
Use only the attached text.

The verbatim phrases become hooks, headlines and ad angles later in this module.

Step 3: draft personas as hypotheses

Using the competitor table and customer themes above, draft 3 audience personas.
For each: situation, main job-to-be-done, top 3 pains (quote customers where possible),
what they've tried, objections, where they spend time, the message most likely to resonate.
Label each persona as a HYPOTHESIS and list 3 questions to validate it with real customers.

AI personas are a starting point, not truth. Validate with a handful of customer conversations, a poll, or your analytics.

Worked example (illustrative)

Omar runs marketing for a meal-prep startup in Riyadh. His deep research scan finds five competitors, all promising "healthy and convenient", with similar price ranges; two emphasize fitness, none emphasizes family meals. Mining 400 anonymized app reviews shows the strongest complaint theme is "same meals every week", and customers repeatedly say "I just don't want to think about dinner". His three hypothesis personas include "working parents who want family dinners handled". Five customer calls confirm it. The new positioning ("Family dinners, decided for you, with a new menu every week") comes straight from research that took one afternoon.

Hands-on: your one-afternoon research sprint

  1. Write the research brief for one real decision (30 minutes including the deep research run).
  2. Spot-check five sources (15 minutes).
  3. Export and anonymize 100 to 500 reviews or survey answers; run the mining prompt (30 minutes).
  4. Draft hypothesis personas and write the validation questions (20 minutes).
  5. Book three short customer conversations to test them.

Before: "Tell me about my target audience" returns a generic persona called "Busy Sarah, 34, loves yoga".

After: a competitor table with links, six customer themes with verbatim quotes, and three labeled hypotheses tied to evidence and ready to validate.

Keep a living research file

Save the outputs (competitor table, customer themes with quotes, hypothesis personas and what you learned from validating them) in your brand project or a shared doc. Re-run the market scan every quarter and the review mining monthly; new themes are often the earliest warning of a product or service problem.

Pitfalls

  • Treating AI personas as research findings rather than hypotheses.
  • Pasting reviews with customer names or order numbers.
  • Trusting market-size numbers without a primary source; if none exists, say so.
  • Researching private individuals; stick to companies, markets and public, professional information.

How to measure success

Research is working when it changes a decision: a new positioning, a message that tests better, or an idea you drop early. Track which campaign messages came from customer phrases and how they perform against your usual copy.

Video lecture: Market and audience research with AI

Lecture coming soon · 11 chapters · about 8 minutes. Read the full transcript below.

  1. Market and audience research with AI
  2. Why research is the big win
  3. The detective analogy
  4. Job 1: the market scan
  5. Job 2: voice of the customer
  6. Job 3: hypothesis personas
  7. Example 1: the review goldmine
  8. Example 2: the meal-prep positioning
  9. Watch me do it
  10. Common mistakes
  11. Recap and try this now

Lecture transcript

Market and audience research with AI

How long did your last competitor analysis take? If you did it properly, reading websites, app store pages, reviews and news, probably days. And if you didn't have days, you probably skipped it and went with your gut. AI changes that math completely. In one afternoon you can run a cited competitor scan, mine hundreds of your own customer reviews for the exact words they use, and draft personas to test. In this lecture you'll learn the three research jobs, the prompts for each, and the checks that keep it honest. You'll see a simple example and a real positioning decision, and watch me mine customer reviews live.

Why research is the big win

Why start the workflows module with research? Because everything downstream depends on it. Your briefs, your captions, your ads, your landing pages all get better when they're built on real competitor context and real customer language. And research is where AI saves marketers the most time, because it's reading heavy work that humans find slow and tedious. But here's the key idea. AI research is only as good as three things: the brief you give it, the sources it uses, and the checks you do afterward. Get those right and you'll make better decisions, faster. Get them wrong and you'll build a campaign on confident guesses.

The detective analogy

Think like a detective. A good detective starts with a clear case file: what exactly are we trying to find out, and why? That's your brief. Then they interview witnesses, and they know some witnesses are more reliable than others. Those are your sources: official statistics and company websites are strong witnesses, anonymous forums and top ten listicles are weak ones. And when they form a theory, they write it on a sticky note as a hunch, not a conclusion, until the evidence confirms it. That's exactly how we'll treat AI personas: as hypotheses to test, not findings to trust.

Job 1: the market scan

Job one is the market and competitor scan, and the right tool is deep research, in ChatGPT, Claude, Gemini or Copilot, or Perplexity. The brief matters. State the decision it informs, like how to position our new meal prep service in Riyadh. List the questions: who the main competitors are, how they position, what customers praise and complain about. Set the scope: which cities, and sources from the last eighteen months. Name your trusted sources, and tell it to treat anonymous forums and vendor blogs as low confidence. And ask for a competitor table with a source link on every row, plus a section on what it couldn't verify. Then spot check five links.

Job 2: voice of the customer

Job two is voice of the customer mining, and it's the most underrated research you can do, because nobody else has your customers' words. Export your reviews, survey answers or support messages, remove names and order numbers, and upload them. Ask for the top themes, praise and complaints separately, with rough counts. Ask for three verbatim customer phrases per theme. And ask for the words customers use for the problem you solve, in their language, not yours. Those verbatim phrases are gold. Later in this module, they'll become hooks, headlines and ad angles that sound like your customers, because they are your customers.

Job 3: hypothesis personas

Job three is personas. Once you have the competitor table and the customer themes, ask your assistant to draft three audience personas: their situation, the job they're trying to get done, their top pains in customer quotes, what they've tried, their objections, where they spend time, and the message most likely to resonate. And add two instructions that make all the difference. Label each persona as a hypothesis. And list three questions to validate it with real customers. Then actually validate them, with a few customer calls, a poll, or your analytics. An AI persona you've tested is a strategy asset. An untested one is fiction with a stock photo.

Example 1: the review goldmine

A simple example. A small candle shop in Manchester exports two hundred reviews and runs the mining prompt. The owner expected the top theme to be long burn time, which is what her product page emphasizes. Instead, the biggest praise theme is nostalgia, with phrases like smells like my grandmother's house and reminds me of winter at home. That's a campaign angle she'd never have written herself. She tests a product page headline and an email subject line built on that language. It's her customers' own words, so it resonates, and it took about twenty minutes to find.

Example 2: the meal-prep positioning

Now a realistic scenario, with illustrative details. Omar runs marketing for a meal prep startup in Riyadh. His deep research scan finds five competitors, and every one promises healthy and convenient, at similar prices. Two focus on fitness. None talks about family meals. Then he mines four hundred anonymized app reviews. The strongest complaint theme is same meals every week, and customers keep saying, I just don't want to think about dinner. His hypothesis personas include working parents who want family dinners handled. He validates it with five customer calls, and four of the five light up at the idea. The new positioning: family dinners, decided for you, with a new menu every week. One afternoon of research, and a genuinely different position in a crowded market.

Watch me do it

Let me mine some reviews. I've exported three hundred reviews for a home fragrance brand and removed names and order numbers. I attach the file and paste the prompt: group into the top six themes, praise and complaints separately, with rough counts. Three verbatim phrases per theme. The words customers use for the problem we solve. And anything surprising. Use only the attached text. Here's the result. Top praise: scent strength, packaging as a gift. Top complaint: diffuser sticks arriving broken. Surprising: several customers mention buying it for a new home. Now, because I want to be sure the quotes are real, I search the file for three of them. All three are there, word for word. The new home angle goes on my list for next month's campaign.

Common mistakes

The common mistakes. Treating AI personas as research findings instead of hypotheses. Uploading reviews or tickets that still contain customer names, emails or order numbers. Trusting market size numbers without a primary source. If the report says the market is worth a precise figure, ask where that comes from, and if there's no reliable source, say so in your deck rather than quoting it. And researching private individuals. Keep research to companies, markets and public, professional information. It's more ethical, it's what privacy law expects, and frankly it's more useful.

Recap and try this now

Let's recap. Use deep research for cited competitor scans, with a brief that states the decision, the scope, trusted sources and what it couldn't verify, and spot check five links. Mine your own reviews, surveys and support messages for themes and verbatim customer language. And draft personas as hypotheses, with validation questions, then test them. Here's your try this now. Block one afternoon for the research sprint in the lesson text: one brief, one deep research run, one review mining session, three hypothesis personas and three customer conversations booked. It's the highest return afternoon you'll spend this quarter.

Key takeaways

  • Use deep research for cited competitor scans, file analysis for your own customers' words, and an assistant for hypothesis personas.
  • Brief research with the decision, scope, time frame, trusted sources and an explicit 'what you couldn't verify' section.
  • Voice-of-customer mining gives you verbatim phrases that become hooks, headlines and ad angles.
  • Label AI personas as hypotheses and validate them with real customers or data.

Try it

Run a one-afternoon research sprint: one deep research brief, review mining on 100-500 anonymized reviews, and three hypothesis personas with validation questions.