Skip to content
AI stack, job

AI for market and customer research

Last verified

Use AI for research to collect and sort what customers and competitors already said, not to invent insights. Point the agent at reviews, forums and ad libraries through a scraping tool, ask for verbatim quotes with links, then check a sample by hand before any line becomes an ad hook.

The workflow for this job
The workflow for this jobFour steps in a loop: the model proposes, the output is checked, a person approves, then it ships and is measured, and what you learn goes back into the next brief.01 MODEL PROPOSESClusters withquotes02 CHECKEvery quote has alink03 PERSON APPROVESHand-checks asample04 SHIP AND MEASUREFeeds briefs andtestsNEXT BRIEFThe workflow for this jobFour steps in a loop: the model proposes, the output is checked, a person approves, then it ships and is measured, and what you learn goes back into the next brief.01 MODEL PROPOSESClusters with quotes02 CHECKEvery quote has a link03 PERSON APPROVESHand-checks a sample04 SHIP AND MEASUREFeeds briefs and tests
Illustrative flow, simplified.
The model proposes, a check catches what it got wrong, a person approves, and the result ships and is measured. What you learn feeds the next brief.

What AI is good at here: collect, cluster, quote

AI market research is the use of a language model to gather, sort and quote text that customers and competitors already published. The model reads reviews, forum threads and ad copy faster than you can. The evidence stays in the source.

  • Collect: pull reviews, Reddit threads and ad library entries through a scraping tool.
  • Cluster: group the text by complaint, desire, objection or phrase.
  • Quote: return the exact customer wording with a link to where it appeared.

A model asked for "the top five pain points of buyers in your niche" with no source text will produce plausible sentences from its training data. Those sentences are not research. Counts and percentages also need to come from your own tally of the collected rows, not from the model's summary.

Workflow: sources, scraping tool, quotes with links, hand check

  1. 1.Write the question and pick sources

    State one decision the research feeds, such as hooks for a landing page or positioning against two competitors. Then list three to five sources: app store or product reviews, relevant subreddits, competitor ad library entries, support threads.

  2. 2.Collect through a scraping tool

    Connect an MCP server or a skill that fetches pages and returns clean text with the URL attached. Save the raw output to a file. The model works from that file, not from memory.

  3. 3.Extract verbatim quotes with links

    Ask for exact substrings from the saved text, each with its source URL and a theme label. Reject any row where the quote is a paraphrase.

  4. 4.Cluster and count

    Group quotes by theme, then count rows per theme with a script or a spreadsheet. Treat the count as a count of what you collected, not of the market.

  5. 5.Check a sample by hand

    Open the links for a random sample of quotes and confirm each quote appears in the source, says what the label claims, and comes from a real customer rather than a seller or a bot.

Illustrative example: the agent returns 200 quotes. You open 20 chosen at random and two do not match their source, which is 2 out of 20, or 10%. Re-run the extraction with a stricter prompt and check a fresh sample. The 10% trigger is a working rule you set, not a standard.

Hand-check rules before a quote becomes an ad hook
CheckPass ruleIf it fails
Quote exists in sourceThe exact words appear on the linked pageDrop the row and flag the prompt
Link opensThe URL loads and shows the same thread or reviewDrop the row
Theme label fitsA second reader would give the same labelRelabel by hand
Author is a customerNot the brand, a competitor or an obvious promoterDrop the row
Date is currentThe quote is recent enough for the product as it is nowMark as dated or drop

Picking a model for extraction and clustering

No clear leader exists for research jobs, and this page ran no model comparison. Choose by job and cost: a cheap tier for tagging and extraction at volume, a stronger tier for reading clusters and drafting positioning. Prices in the table are per million tokens, were checked on 15 November 2026, and some carry end dates, so confirm them on the vendor page before you budget.

Models for market and customer research
NameVendorJobStatusPrice as the vendor states itCheckedSource
Claude Fable 5.2claude-fable-5-2.AnthropicAd copy, Market research, Analytics, Paid UA agentsCurrent$8 input / $40 output per MTok.AnthropicAnthropic
Claude Opus 5.5claude-opus-5-5.AnthropicAd copy, Market research, Analytics, Paid UA agentsCurrent$4 input / $20 output per MTok.AnthropicAnthropic
Claude Sonnet 5.5claude-sonnet-5-5.AnthropicAd copy, Market research, Analytics, Paid UA agentsCurrent$2 input / $10 output per MTok.AnthropicAnthropic
Claude Haiku 5.5claude-haiku-5-5.AnthropicAd copy, Market research, Analytics, Paid UA agentsCurrentFrom $0.10 input / from $0.50 output per MTok.AnthropicAnthropic
GPT-6 Astragpt-6-astraOpenAIAd copy, Market research, Analytics, Paid UA agentsCurrent$10 input / $50 output per MTok (standard, short context).OpenAIOpenAI
GPT-6.1 Solgpt-6.1-solOpenAIAd copy, Market research, Analytics, Paid UA agentsCurrent$2 input / $10 output per MTok.OpenAIOpenAI
GPT-6.1 Terragpt-6.1-terraOpenAIAd copy, Market research, AnalyticsCurrent$1 input / $5 output per MTok.OpenAIOpenAI
GPT-6 Lunagpt-6-lunaOpenAIAd copy, Market research, AnalyticsCurrent$0.10 input / $0.50 output per MTok.OpenAIOpenAI
Gemini 4 Carbongemini-4-carbon.GoogleAd copy, Market research, Analytics, Paid UA agentsCurrent$2 in / $10 out per MTok until 2026-11-30, then $4 / $20.Google, stated until 30 November 2026Google
Gemini 4 Argongemini-4-argon.GoogleAd copy, Market research, Analytics, Paid UA agentsCurrent$2 in / $10 out per MTok, rising to $4 / $20 later per page.GoogleGoogle
Gemini 4 Hydrogengemini-4-hydrogen.GoogleAd copy, Market research, Analytics, Paid UA agentsCurrent$1.50 in / $7.50 out per MTok.GoogleGoogle
Gemini 3.8 Flashgemini-3-8-flash.GoogleAd copy, Market research, Analytics, Paid UA agentsCurrent$0.75 in / $3.75 out per MTok until 2026-12-31, then $1.50 / $7.50.Google, stated until 31 December 2026Google
Grok 4.7grok-4.7xAIAd copy, Market research, Analytics, Paid UA agentsCurrent$2 in / $0.50 cached / $6 out per MTok (up to 200k prompt), $4 / $12 above. 500k context, knowledge cutoff May 2026.xAIxAI
Muse Spark 1.1muse-spark-1.1MetaAd copy, Market research, AnalyticsPreviewSee vendordev.meta.ai
Advantage+ creative generative AI features in Ads ManagerMetaStatic creatives, Ad copy, Market researchCurrentSee vendorabout.fb.com
Qwen models (Model Studio)AlibabaAd copy, Market research, Static creatives, AnalyticsCurrentSee vendorAlibaba
JevTypeSafe AIAd copy, Market research, AnalyticsLimited access$42 per billion input tokens ($0.042 per million), input tokens only; output tokens free (currently).TypeSafe AITypeSafe AI

Prompt template

Three short prompts cover the loop. Paste your saved source file where the template says so. Keep the rules about quotes in the system prompt so they apply to every run.

  1. 1.System rules

    Set the constraints once.

    Use only the text in SOURCE_FILE. Return exact substrings as quotes. Attach the source URL to every quote. If the text does not support a claim, write NO EVIDENCE. Do not give counts; the user counts rows.
  2. 2.Extraction prompt

    Ask for rows you can check and count.

    Question: [one decision, one audience]
    From SOURCE_FILE return JSON rows: {"quote": exact substring, "url": source link, "theme": short label, "emotion": one word}. Skip rows where the author sells the product.
  3. 3.Verification prompt

    Run on a second pass, then still open the sample by hand.

    For each row, check that quote is an exact substring of the text at url in SOURCE_FILE. Output PASS or FAIL per row with a one-line reason. Do not edit any quote.

Data leaving the machine and terms of the sources

Scraping tools send URLs, and sometimes page text, to a third-party service. Reviews and forum posts often include usernames and personal details. Strip handles before the text reaches a model, and keep customer lists and internal documents out of the research folder.

  • Read the terms and robots rules of each source before you collect from it. This page does not say that scraping any site is allowed.
  • Ad libraries and review sites each set their own access rules, and those rules change.
  • Check where each tool sends data: some MCP servers send URLs to the vendor's cloud, and some skills call a separate service with your API key.
  • Keep keys in environment files, not in prompts or saved transcripts.

Skills and MCP for research

Skills are instruction packs that can bundle scripts. MCP servers are the connectors that fetch the data. The tables list repositories for this job with licence and scan notes from 15 November 2026. Treat each one as unreviewed third-party code and read its licence yourself, because terms differ.

Skills for this job
NameJobLicenceChecked
appeeky/aso-skillsASO, Market researchMITClass A
aaron-he-zhu/aaron-marketing-skillsSEO and GEO, Market researchApache-2.0Class A
firecrawl/firecrawl-workflowsSEO and GEO, Market researchISCClass A
PostPlusAI/postplus-skillsPaid UA agents, Market research, Social contentPolyForm-Shield-1.0.0Class B
lignertys/reddit-research-skillsMarket researchMITClass B

Firecrawl and Apify are the scraping options in the MCP table. Firecrawl sends URLs to its cloud. The Apify server runs third-party scrapers (Actors) with your Apify token, so review each Actor before you point it at an ad library or social source.

MCP servers for this job
NameJobPublisherAccessLicenceChecked
apify/apify-mcp-serverPaid UA agents, Market research, Social contentPlatform vendorNot reviewedMITClass A
firecrawl/firecrawl-mcp-serverMarket researchPlatform vendorNot reviewedMITClass A
KenanAtmaca/aso-mcpASO, Market researchThird partyNot reviewedMITClass B

What this page does not claim

  • It ranks no model for research. No test on research tasks was run here.
  • It gives no install counts, star counts or vendor rankings for the repositories listed.
  • It does not endorse any third-party skill or server, and none is official or partnered with this site.
  • It reports no results from a research project, only a method you can check.

Questions

Which AI model fits market research?
No single model is shown to lead here. Match the model to the job: a cheap tier for extraction and tagging, a stronger tier for reading clusters. Run both on the same saved source file, then compare the quotes that pass your hand check, and compare cost.
Can AI replace customer interviews?
AI reads what customers already wrote, so it covers public language well. It cannot ask follow-up questions or reach people who never posted. Use it to find themes and wording, then confirm the important ones with real conversations.
How do you stop an AI from inventing customer quotes?
Give the model a saved source file, require exact substrings with the source URL, and allow NO EVIDENCE as an answer. Then open a random sample of links yourself. A model check helps, but your sample is the final test.
Can you scrape Reddit and ad libraries for research?
That depends on each source's terms, robots rules and your jurisdiction, and this page does not confirm that any site allows it. Read the terms first, prefer official access where it exists, and collect only what the research question needs.

For a worked example of tagging a whole ad market, see Claude + Jev on 54 live ads: descriptions, typed tags, a counted snapshot and a review queue.

See the real run

Sources

Each source shows the date it was last checked.

Show all 26 sources
  1. about.fb.com about.fb.com/news/2025/02/gen-ai-transparency-metas-ads-products/
  2. Meta ai.meta.com/blog/muse-spark-1-1/
  3. Google aistudio.google.com/docs/models
  4. Google aistudio.google.com/docs/pricing
  5. Alibaba alibabacloud.com/help/en/model-studio/models
  6. Anthropic claude.com/pricing
  7. dev.meta.ai dev.meta.ai/resources/blog/build-with-muse-spark/
  8. OpenAI developers.openai.com/api/docs/models
  9. OpenAI developers.openai.com/api/docs/pricing
  10. TypeSafe AI docs.typesafe.ai/api/system-one.md
  11. TypeSafe AI docs.typesafe.ai/introduction/quickstart.md
  12. TypeSafe AI docs.typesafe.ai/models.md
  13. xAI docs.x.ai/docs/models
  14. Meta facebook.com/business/news/Introducing-Enhanced-Gen-AI-Features-and-Other-Tools-to-Help-Build-Your-Business
  15. GitHub github.com/aaron-he-zhu/aaron-marketing-skills
  16. GitHub github.com/apify/apify-mcp-server
  17. GitHub github.com/appeeky/aso-skills
  18. GitHub github.com/firecrawl/firecrawl-mcp-server
  19. GitHub github.com/firecrawl/firecrawl-workflows
  20. GitHub github.com/KenanAtmaca/aso-mcp
  21. GitHub github.com/lignertys/reddit-research-skills
  22. GitHub github.com/PostPlusAI/postplus-skills
  23. learn.chatgpt.com learn.chatgpt.com/docs/pricing
  24. Anthropic platform.claude.com/docs/en/about-claude/pricing
  25. Anthropic platform.claude.com/docs/en/about-claude/models/overview
  26. TypeSafe AI typesafe.ai

Want your own AI and acquisition setup checked for gaps and risks? Start with a growth audit.

See the growth audit