- What I use, week to week
- A story about a quote that never existed
- A step-by-step process that catches the fake stuff
- Where the free tools stop being enough
- Don't skip the humans
- Who should be running this, day to day
- The one thing to check before you trust any AI research output
- Frequently asked questions
- Official documentation
The short version: Perplexity and ChatGPT with Deep Research are the best starting points for market and competitor research, SparkToro and AnswerThePublic are still unmatched for audience insight, and Claude is the one I trust most for pulling apart long reports without making things up. None of them replace a human checking the sources before you publish anything with a number in it.
What I use, week to week
I run marketing research most weeks for my own business and for clients, and the tool stack changes roughly every quarter because someone always ships something better. Here's what's on my desktop right now, and why.
Perplexity, for the first pass
Perplexity is my default for "what's happening in this market" questions because it shows its sources next to the answer, which means I can click through and check them in under a minute. I used it last month to pull together a snapshot of the UK coaching industry for a client pitch, ten data points in fifteen minutes, versus the two hours that would have taken me in 2019 with a spreadsheet and eight open browser tabs.
ChatGPT with Deep Research, for the long version
When I need a proper competitor teardown, ten competitors, positioning, pricing tiers, review sentiment, I use ChatGPT's Deep Research mode and let it run for eight to twelve minutes while I do something else. It comes back with a structured report and citations. It is one of the more useful things ChatGPT does for business right now, alongside drafting and summarising, which I've written about in more detail on what AI tools like ChatGPT are used for in business.
Claude, for reading the boring stuff
Claude is what I feed a 40-page industry report or a client's last three years of customer support tickets. It's less prone to inventing a statistic than ChatGPT when the source document is long and dense, and it's better at saying "this document doesn't mention that" instead of guessing.
SparkToro, for finding out who listens to whom
SparkToro tells you which podcasts, YouTube channels, newsletters and Twitter/X accounts your audience follows. It's not built on an AI model that generates opinions, it's built on real audience data, which is exactly why it's still one of the few tools in this space I never question. If you're planning outreach or a partnership, this is the one to open before you open ChatGPT.
AnswerThePublic and AlsoAsked, for what people are typing
These scrape real search queries and autocomplete data, so you get the actual questions people are asking, not an AI's guess at what they might ask. I pair this with basic keyword volume work, and if you're wondering how many terms you need to build a content plan around, I go through that in how many keywords you really need for content research.
Otter.ai and Grain, for turning conversations into data
If you run customer interviews (and you should, more on that below), Otter or Grain transcribe and timestamp them so you can search "pricing" across forty interviews instead of re-listening to hours of audio. This is unglamorous and it's the single biggest time saver on this list.
A story about a quote that never existed
Two years ago I asked ChatGPT to pull together customer sentiment on a competitor's product for a positioning document. It came back with a clean, specific line: a supposed customer review saying the onboarding was "confusing and took three attempts to set up." It read exactly like a real review. It sounded like something a person would say.
It wasn't real. I searched for the exact phrase, on the competitor's site, on Trustpilot, on Reddit, nowhere. The model had built a plausible-sounding quote because I'd asked for one and it wanted to give me something useful-shaped. If that document had gone into a client pitch deck as evidence, and the client had repeated it in a sales call, that's a fabricated claim about a named competitor sitting in someone's mouth.
This is the bit most people writing about AI research tools skip over, because it's not a flattering story about the technology. AI research tools are fast at synthesis and bad at telling the difference between "this is true" and "this sounds true." The faster and more polished the output looks, the less most people check it, and that's backwards. Polish should make you more suspicious, not less.
A step-by-step process that catches the fake stuff
Here's the actual workflow I use now for any research that's going in front of a client or a public post:
- Ask the tool for the claim and the source in the same prompt, every time, no exceptions.
- Click through to the source before the number leaves your document. If there's no link, treat the number as invented until proven otherwise.
- Cross-check any statistic with one other source, ideally Google Trends, a published industry report, or a government data set.
- Run competitor or customer language claims through a plain text search on the actual platform (Trustpilot, G2, Reddit, App Store reviews) rather than trusting a summary of them.
- Keep a running note of which tool got which fact wrong. Perplexity and Deep Research get better over time in aggregate, but they don't remember your specific corrections, so you're the one holding the memory.
That five-step check adds maybe fifteen minutes to a research task. It has saved me from publishing at least three wrong numbers in the last year alone.
Where the free tools stop being enough
Google Trends, AnswerThePublic's free tier and Perplexity's basic mode will get a solo founder or a small content team most of the way. Where you start needing paid tools is competitive keyword and backlink data, which is where SEMrush (from around £99 a month) or Ahrefs (from around $99 a month) come in, and neither has been meaningfully replaced by AI chat tools because they're built on crawled web data rather than a language model guessing. If you're trying to work out what a realistic monthly spend looks like across a small marketing stack, I've broken the real numbers down in how much AI tools cost for small teams, and it's more than most people budget for once you add up three or four subscriptions rather than just the one.
Want AI doing the heavy lifting in your marketing?
I build the systems that handle the boring 80 percent, so you get your week back. Done properly, with the human kept in.
Don't skip the humans
None of this replaces talking to actual customers. I still run five to eight customer interviews before any big positioning project, thirty minutes each, recorded, transcribed with Otter, then themed in Dovetail or manually if the project is small. AI tools are brilliant at telling you what the market is saying in aggregate. They can't tell you why your specific customer nearly cancelled last Tuesday. That still requires a phone call.
This matters more than it sounds like it should, because marketing automation as a category has always promised to remove the human step and never quite managed it, going back to the earliest lead-scoring tools in the 1990s. I go into that history in when marketing automation began and why it still matters, and the short version is that every wave of "automate the research" tools has still needed a person checking the output before it went anywhere near a customer.
Who should be running this, day to day
If you're hiring for this rather than doing it yourself, the job title to look for is usually content marketing specialist or marketing researcher, and the pay reflects how much judgement it takes to run these tools well rather than just prompt them. I've laid out realistic ranges in how much a content marketing specialist makes, and the short answer is that someone good at this, who can tell a real insight from an AI-generated pattern that isn't there, earns more than the job title suggests.
The one thing to check before you trust any AI research output
Whether it's for a blog post, an academic-style paper, or a client report, the standard is the same: check whether the tool is generating the content or generating the leads to the content. Generated summaries with no traceable source are the risk category. I've written a longer answer to this exact question, framed around research papers but the rule applies everywhere, in how much AI writing is acceptable in a research paper. If you can't trace a claim back to a primary source in under two minutes, don't publish it as fact.
See also The AI Tools I Use to Run My Marketing Business and What Each Replaced, which picks up where this leaves off.
If you write about research or data, see my live-checked list of science blogs that accept guest posts.
Frequently asked questions
What is the single best AI tool for marketing research right now?
For most people it's Perplexity, because it shows sources next to the answer so you can verify claims quickly, but pair it with SparkToro for audience data since Perplexity can't tell you who your customers listen to.
Can I trust the statistics an AI tool gives me for a client report?
Not without checking. AI models sometimes invent plausible-sounding statistics and quotes when asked directly for evidence, so always ask for the source, click through, and confirm the number exists before it goes into anything client-facing.
Do AI research tools replace customer interviews?
No. They're strong at summarising public data and competitor positioning but can't explain why one specific customer nearly churned or what made them buy. Run five to eight real customer interviews for any major decision alongside the AI research, not instead of it.
How much should a small business expect to spend on AI research tools monthly?
A realistic small-team stack, one AI chat subscription, one SEO tool like SEMrush or Ahrefs, and a transcription tool, usually lands between £150 and £300 a month once you add the paid tiers together, rather than the single subscription price most people budget for.
