ChatGPT, Claude, Gemini, Perplexity and Microsoft Copilot can all research competitors, prices and demand from the web. Individual plans cost about $20 a month each, Copilot's business licence $21 a user, and Gemini's free tier and Google Trends nothing. None replaces checking: open every source the AI cites and confirm the number is on the page.
The cost that matters isn't the subscription. It's the time you spend verifying, because these tools write fluent paragraphs with figures that sound right whether or not a source supports them. OpenAI's own announcement of deep research says it can sometimes hallucinate facts or make incorrect inferences, may struggle to tell authoritative information from rumours, and often fails to convey uncertainty accurately. Budget about ten minutes for each claim that would change a price, a budget or a promise to a customer. A typical small project has around ten such claims, so the checking comes to under two hours.
Four questions a small business needs answered, not a market size
Global market-size figures make impressive slides and rarely change a decision. A small business needs four local answers, and each points to a different tool.
- Who else sells this to my customers, and what exactly do they offer? That is competitor research. It needs web research that reads specific pages.
- What do they charge, and for what? That is pricing research. It needs a table with units and dates, which is where most AI output goes wrong.
- When do customers look for this, and is interest rising? That is demand research. It needs search-trend data, not a paragraph.
- What do customers praise and complain about? That is customer-voice research. It needs a pile of public reviews and an assistant to group them.
A fifth job sits beside these: analysing your own sales or enquiry data. It needs an assistant that can read your files, not one that searches the web. Each tool below suits some of these jobs better than others. For the two most often compared, Perplexity versus ChatGPT for business research goes into more depth.
Tools compared: costs, what each does and where it is weak
Prices are USD list prices as of October 2026, from each vendor's own page. Confirm before you buy, because plans and allowances change.
| Tool | Plan and list price | What it does for market research | Watch for |
|---|---|---|---|
| ChatGPT (deep research) | OpenAI's pricing page lists limited deep research on Free and Go, expanded on Plus ($20 a month) and maximum on Pro (from $100 a month). | Long multi-source reports. OpenAI says a task can take 5 to 30 minutes. | Allowances differ by plan. OpenAI itself warns that it can hallucinate facts and misjudge how certain it is. |
| Claude (Research) | Needs a paid plan: Pro is $20 a month, or $17 a month billed as $200 a year. Team is $25 a seat monthly ($20 annual), minimum two seats. | Runs several searches that build on each other. Claude can also work from your own files in a Project, which suits the sales-data job. | Web search must be on, and Anthropic says research sessions use your limits faster than ordinary chats. |
| Gemini (Deep Research) | Listed on the free plan, with limits Google doesn't publish as numbers. Google AI Pro is $19.99 a month with higher limits. | Drafts a research plan you can edit before it starts, then writes a report in about 5 to 10 minutes. Can include your Drive or Gmail files. | Daily caps aren't published as numbers, and the app only says how many research requests are left when you are close to the limit. |
| Perplexity (Pro) | About $20 a month on monthly billing; Perplexity's Pro page shows $17 a month billed annually and lists deep research as a Pro feature. | Fast answers. Perplexity says every answer includes inline citations so you can verify claims. | Inline citations make an answer look checked. They don't make it correct. |
| Microsoft Copilot (Researcher agent) | Paid Microsoft 365 Copilot licence: $21 a user a month on annual billing for Copilot Business ($25.20 monthly), on top of a Microsoft 365 plan. A promotional $18 runs to 31 December 2026. | Combines the web with your files, email and meetings, and returns a cited report. Free Copilot Chat is web-grounded but lists advanced agents as limited. | Only worth it if you already live in Microsoft 365. Check which research features your licence includes. |
| Google Trends | Free | Relative search interest over time, with up to 8 groups of terms compared at once on the current Explore page (5 on the classic one). Good for seasonality. | Values are scaled 0 to 100 from a sample of searches, not volumes. |
| Google Keyword Planner | Free, but it sits inside Google Ads: you need an account with billing details entered before the basic features open | Estimates of monthly searches for a phrase. | Low-volume and sensitive keywords aren't shown. What an account without ad spend sees is unconfirmed, so check yours. |
| Gemini Notebook (formerly NotebookLM) | Free: 50 sources a notebook and 50 chats a day. Paid Google AI plans raise this (300 sources, 500 chats on Pro). | Works only from documents you add, such as saved competitor pages and price lists, which narrows what it can invent. | It can't see anything you didn't add, so gaps in your sources become gaps in the answer. |
I've left out paid SEO and market-intelligence suites. Their plans and limits change often, and I couldn't confirm current USD list prices for them from the vendors' own pages, so ask any vendor for a trial and a written price before paying. For a shop or a firm of a handful of people, the tools above answer the four questions without them.
What $0, about $20 and about $40 a month buy
| Stack | Tools | Monthly cost | What you can do | Where it runs out |
|---|---|---|---|---|
| Free | Gemini free tier, ChatGPT's limited deep research, Google Trends, Gemini Notebook free | $0 | One focused competitor scan, a seasonality check and a review summary | Daily caps. Several research runs in one week may stall. |
| One paid plan | Any one of ChatGPT Plus, Claude Pro, Google AI Pro or Perplexity Pro, plus Trends | About $20 | Weekly research runs and analysis of your own files | Each plan's research allowance, and your habit of checking |
| Two tools, two jobs | A web-research plan plus a second assistant for your own files, plus Trends | About $40 | Research one tool, analysis in the other, with a second opinion on sources | The risk of two tools giving two confident answers; the checks decide between them |
The free stack is enough for most one-off decisions. Move to a paid plan when you hit a daily cap in the middle of a project, or when you find yourself researching every month. Buy a plan for the project, put a reminder in the diary to cancel it, and judge it on whether it saved checking time, not on whether the reports looked impressive.
A driving school tests an intensive course: the research, step by step
A four-instructor driving school wants to know whether to offer a one-week intensive course. The questions are who else offers one nearby, at what price and what the price includes, whether interest is seasonal, and what customers say about rivals. The project below takes under three hours of the owner's time and runs on free tiers, with one month of a $20 plan as a fallback if a cap bites. Waiting time for the research run isn't counted.
| Step | Your minutes |
|---|---|
| List eight competitors from the map results, with their course page addresses | 15 |
| Write and run the constrained research prompt (5 to 30 minutes of waiting) | 10 |
| Read the output and flag doubtful rows | 20 |
| Check the facts that would change a price: three per school, eight schools, about two minutes each (quicker than the full routine below, because each fact sits on the school's own page) | 48 |
| Google Trends comparison, with notes | 20 |
| Group 40 competitor reviews by theme with an assistant, then check the counts | 30 |
| Write the one-page recommendation | 30 |
| Total | 173 (about 2 h 53 min) |
The trick that keeps the research honest is the constraint. Instead of asking an assistant to "research the driving school market", give it the eight pages and ban everything else. The wording below works in Gemini's Deep Research, ChatGPT or Claude, and Gemini's version lets you edit the research plan before it starts, which is the right moment to confirm it only uses your list.
You are researching for a small driving school deciding whether to offer a one-week intensive course.
Use ONLY the eight web pages listed below. Do not use any other source.
For each school, report: whether it offers an intensive or fast-track course; the number of lesson hours; the price quoted; what the price includes and excludes (test fee, car, booking fee); the exact sentence on the page that states the price; and the date shown on the page, if any.
If a page doesn't state something, write "not stated". Do not infer or estimate.
PAGES
1. [address of competitor page 1]
...
8. [address of competitor page 8]An illustrative answer, showing three of the eight rows with invented figures:
| School | Intensive offered? | Hours | Price quoted | Includes or excludes | Quote from the page | Page date |
|---|---|---|---|---|---|---|
| A | Yes, "one-week course" | 30 | $1,150 | Test fee not stated | "Our one-week course is from $1,150" | March, 23 months ago |
| B | Yes, "fast-track" | 20 over two weeks | $1,000 | Test fees included | "Includes your theory and practical test fees" | Not stated |
| C | No | Not stated | Not stated | Not stated | Not stated | This year |
Then the checks, which are the point. School A's page was nearly two years old, and the live booking page showed a higher price, so the row was stale. School B's "fast-track" is 20 hours across two weeks, a different product from a one-week intensive, so it can't be compared like for like. School C's "not stated" was correct for the page supplied, but the school has a separate intensive page that wasn't on the list, which is the price of constraining the sources. Of the eight rows, four were usable after checking. The verified band for a comparable 30-hour intensive came out at $1,150 to $1,400 (illustrative), with two of the schools excluding the test fee. The owner's decision is to price inside the band, say what is included, and try two holiday weeks before committing. For a longer version of this exercise, how to do competitor research with AI in one afternoon is the companion tutorial, and running a competitor brief with ChatGPT deep research and a repeatable Gemini Deep Research brief cover each tool's quirks.
The ten-minute source check, claim by claim
You can't check everything, and you don't need to. Check every number or fact that would change a price, a budget or a promise to a customer, and skim the rest. A realistic budget is ten claims a project at about ten minutes each. Checking sources and citations in AI research covers the principles; this is the routine I'd run on market research.
- Pick the claims that matter (2 minutes). Highlight the figures a decision depends on. Ignore the colour commentary.
- Open the cited page itself (1 minute). Use the page, not the AI's summary of it. The claim fails if the link is dead, lands on a homepage, or sits behind a paywall you can't open.
- Find the figure on the page (1 minute). Use the browser's find box. If the number or the sentence isn't there, the citation doesn't support the claim, whatever the report says.
- Check the definition and the unit (2 minutes). What place, what period, what currency, what is included? A price "per hour" and a price "per package" don't compare, and "market" can mean a global category you don't sell in.
- Check the date and the publisher (1 minute). When was it written, and who gains from the claim? A vendor saying its own category is growing is marketing, not evidence.
- Find a second source that doesn't cite the first (3 minutes). For any claim that moves money, this step isn't optional. Two pages repeating one origin count as one source.
- Recompute the arithmetic (1 minute). Percentages, totals and per-unit prices get redone with a calculator, not by asking the same assistant again.
- Log the result (1 minute). Claim, source, what the page says, verdict and confidence. The log is what lets someone else rely on your work in three months.
A source log, filled in: one claim passes, one fails, one changes
Here are three entries from the driving school's project, with an invented market-size claim added to show a typical failure. The log takes about a minute a row once the checking is done.
| Claim | Source cited | What the page says | Verdict | Confidence and use |
|---|---|---|---|---|
| School B offers 20 hours over two weeks for $1,000, test fees included | School B's courses page | The sentence is on the page, found with the find box, dated this year | Pass | High. Use it as a different product, not a direct comparator |
| Interest in intensive courses peaks in the school holidays | Google Trends chart, read directly (the AI had only described it) | Relative interest on a 0 to 100 scale peaks twice a year | Pass with a caveat | Medium. Relative, not volume: use it to schedule courses, not to forecast bookings |
| "The driver training market will grow 9% a year" (an invented example claim) | A market-research firm's summary page | The figure isn't on the visible page, the full report is paywalled, and the page covers global driver-training software | Fail | None. Drop it. It doesn't affect a local pricing decision anyway |
The third row is the one to remember. A tidy market-size sentence from a research firm is often a teaser: a headline figure on a summary page, a paywalled report behind it, and a category that is nearly yours but not quite. A small business can almost always ignore these figures, because bottom-up local counts answer the real question better: how many competitors, what they charge, and when customers look.
Demand numbers: what Trends and Keyword Planner can and can't say
Google's help page for Trends is clear about what the numbers are. Only a sample of searches is used, each value is divided by the total searches for its region and period, and the results are scaled from 0 to 100. So a 100 means the peak for that chart, not 100 searches, and the page advises using relative comparisons. Used that way, Trends is excellent for two jobs: seeing the seasonal shape of demand, and comparing the popularity of two phrases.
A florist who needs to schedule staff and stock might set the range to five years, so repeated seasons show, and compare "flower delivery" with "wedding flowers". The chart (illustrative) shows two sharp peaks a year for one phrase and a long summer plateau for the other. The florist takes two decisions from it: extra hands in the weeks before each peak, and wedding consultations booked into the plateau. They don't take a sales forecast from it, because the chart contains no sales or volumes. Google also says plainly that Trends is not a scientific poll, which is a good habit to bring to any search-interest chart. Google's current Explore page also uses Gemini models; treat anything it suggests like any other AI output and check it against the chart.
Keyword Planner answers a different question: roughly how many searches a phrase gets each month. Google calls it a free resource for advertisers, but it sits inside Google Ads, and its help page says you must finish account set-up, billing details included, before even the basic keyword-ideas feature opens. The same page says it shows estimates, and that keywords with very low volumes or sensitive topics can't be discovered or forecast. I couldn't confirm from Google's pages whether an account that has never spent on ads sees exact numbers or only ranges, so look at what yours shows before you build anything on it.
Competitor pricing: capture it so it survives a second look
Price research fails on units far more often than on facts. Take a photography studio comparing two competitors' mini-session offers. The raw numbers look like this: one charges $120 for a 20-minute session with 8 edited images, the other $165 for 30 minutes with 12 images. Which is cheaper depends on the unit.
| Competitor | Package | Includes | Price | Per minute | Per image | Date seen and address |
|---|---|---|---|---|---|---|
| X | Mini session | 20 minutes, 8 edited images | $120 | $6.00 | $15.00 | [date], [page address], screenshot saved |
| Y | Mini session plus | 30 minutes, 12 edited images | $165 | $5.50 | $13.75 | [date], [page address], screenshot saved |
Y is cheaper on both units, but only because it includes more, and what the studio should do with that depends on what its own customers value. The sheet's job is to make the comparison possible and defensible. Three habits make it hold up: record what each price includes, record the date and address, and save a dated screenshot of every price you rely on, because pages change and a research report can't tell you what a page said last month. Where a page doesn't say, write "not stated". An assistant can build the table and do the per-unit sums quickly, which is worth using it for, but recompute two of them yourself.
Customer voice from reviews: a bookshop's one-hour read
Reviews of other shops in your field tell you what customers wish existed. An independent bookshop could copy 40 public reviews of three nearby competitors into an assistant, with a prompt that asks for themes with counts, not quotes.
Below are 40 public reviews of three bookshops, with names removed. Group what readers praise and what they complain about into themes.
For each theme give the number of reviews that mention it and a short paraphrase (not a quote). Mark any theme mentioned in fewer than 3 reviews as weak.
Do not add themes that no review mentions.Praise: staff recommendations (14 reviews), author events and book clubs (9), children's section (7). Complaints: special orders taking longer than promised (6), parking (4). Weak: gift wrapping (2).
That is an illustrative answer, and the checks are quick. Count one theme by hand (the staff recommendations count is the one to test), because assistants are often off by a few when they tally. Use the output for your own thinking only. Reviews belong to the people who wrote them, so don't quote them in your marketing without asking. If the same complaint appears in your own reviews, that is a shop problem, not a research finding.
Seven red flags in AI market analysis
- A citation that lands on a homepage, or a dead link. The assistant is citing a source it can't point you to.
- A precise figure with no definition. "$4.2bn" of what, where, in which year and which currency? An answer that can't say is not an answer.
- Global numbers for a local decision. They impress and mislead. Replace them with local counts.
- Missing or old dates. Data from several years ago presented as current. Ask for the date on every row.
- A table with no gaps. Real pages leave things unstated. A competitor table with every cell filled should make you suspicious, and "not stated" in a few cells is a sign the tool is behaving.
- A vendor's marketing offered as evidence. A company's own page describing the size of its own category is a claim, not a measurement.
- One number repeated across many sources. They may all trace to the same origin. Three articles quoting one report are one source.
The sentence in OpenAI's announcement that the tool often fails to convey uncertainty accurately is worth keeping in mind as an eighth flag: confident tone tells you nothing. A fluent paragraph with no hedges can be right or wrong, and only the source check tells you which.
Where a person should do the research instead
Some questions are better answered by talking to people than by any tool. If the decision is whether five customers would pay for a new service, ask five customers. If it is what a rival charges for a job that isn't on their website, ask for a quote as a customer would. The cheapest use of an AI research tool is to prepare for those conversations: draft the questions, organise the answers and summarise what you heard. The tools above find and organise public information fast. They don't know your customers, and they can't tell you what a rival didn't publish.
Questions owners ask about AI market research
Is a paid plan worth it for market research?
For a one-off project, try the free tiers first: Gemini lists Deep Research on its free plan and ChatGPT includes a limited allowance. Pay for one plan, about $20 a month, only when you hit limits or research every month, and cancel after the project if you don't. Most of the value comes from the checking routine, not the subscription.
Can I trust the citations an AI research tool gives me?
Treat them as leads, not proof. A cited page can fail to contain the claim, be out of date or repeat a vendor's marketing. OpenAI's own announcement of deep research says it can hallucinate facts and often fails to convey uncertainty. Open the link, find the figure on the page, and note the date and who published it.
How do I compare competitor prices fairly?
Capture the same details for every competitor: what the price includes, the unit (per hour, per package, per image), the date you saw it and the page address. Take a dated screenshot of any price you rely on, because pages change. Write 'not stated' in the cell instead of letting a tool guess.
Does Google Trends show how many people search?
No. Trends scales interest to a range of 0 to 100 and shows relative popularity from a sample of searches, not raw volumes. It's good for seasonality and for comparing two phrases. For estimated monthly searches, Google's Keyword Planner is a separate tool that needs a Google Ads account.
Should I upload customer data to these tools?
Rarely needed for market research, and not before you've read the plan's data terms. Use public information and aggregated figures. If you do analyse your own sales file, use a business plan whose terms say your content isn't used for training by default, and strip names and contact details first.
Further reads
- How to Track Competitors' Prices and Offers With AI — Turn a one-off price capture into a regular competitor tracker.
- How to Test a New Service Idea With AI Before You Launch It — Test a service idea with customers before you commit.
- Is Perplexity Pro Worth Paying For in a Small Business? — Whether the Perplexity subscription earns its price.
- How to Use ChatGPT for Market Research on Your Local Customers — Research your own local customers with ChatGPT.
- AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.
Sources: Anthropic's Claude pricing page and Research help article, OpenAI's ChatGPT pricing page and deep research announcement, Google's AI plans page, Gemini Deep Research help page, Google Trends help pages, Google Ads Keyword Planner help and product pages, Perplexity's Pro page, Microsoft Learn (Researcher agent, Copilot Chat overview), all checked 28 October 2026. Worked examples and invented claims are illustrations, not data.