How to Do Competitor Research With AI in One Afternoon

Pick three to five competitors your customers genuinely compare you with, then work through five timed blocks over about four hours: collect their pages, reviews and ads; have an AI assistant profile each one against the same template; mine their reviews for complaints; compare everyone in one grid; and finish by choosing two or three changes to make.

The afternoon works because AI is quick at reading and sorting dozens of pages, not because it knows your market. Assistants with web search regularly return prices, opening hours and menus that are months out of date, and none of them can see what happens at a rival's counter. So you feed them sources you have checked yourself, you verify anything you plan to act on, and the output is one page of decisions rather than a 30-page report nobody reopens.

Follow me on Instagram@sagnikteaches

The four-hour plan, block by block

Here is the whole afternoon on one screen. The times assume four rivals; add roughly 40 minutes for each extra one, which is why five is a sensible ceiling.

Connect on LinkedInSagnik Bhattacharya

For an illustration, take a 30-seat café with two baristas, whose owner has just heard that a chain coffee shop is opening four doors down. She blocks out a Tuesday from 1pm to 5pm, closes her laptop's other tabs, and treats it like a meeting she can't cancel.

Subscribe on YouTube@codingliquids

Prep: pick the rivals customers actually weigh you against

The competitors that matter are the ones a customer considers for the same job, not every business in your category. Three quick ways to find them:

• Search as a customer would. Open a private browser window, search Google Maps with the words a customer would type ("flat white", "brunch", "coffee open early"), and note who appears in the first screen of results.
• Ask the people who chose you. "Who else did you look at?" is the most useful question in any enquiry form or conversation. A week of asking gives you a real list.
• Include one substitute. A substitute solves the same need in a different way. For takeaway coffee, the supermarket's self-serve machine is a substitute; for a dog groomer, it's the owner with a pair of clippers and a video.

The café's final list: the incoming chain, an independent café 300 metres away with a strong weekend brunch trade, the supermarket bakery counter (the substitute for coffee and a pastry on the go), and a coffee van that parks by the station on weekday mornings.

A bicycle repair shop doing the same exercise would end up with a different kind of list: a mobile mechanic who comes to customers' homes, the service desk inside a large sports retailer, and a community bike workshop with self-service repair stands. None of them look like a traditional bike shop, which is exactly why they're easy to miss.

Block 1: build a source pack you can trust (45 minutes)

If you simply ask an assistant to "research my competitors", it picks its own sources and you can't tell which are stale. Collect the sources first, so every later answer traces back to something you have seen with your own eyes. For each rival, save:

• Their website pages that matter: home page, menu or price list, about page, any "offers" page. Print to PDF or paste the text into a document, and put the date in the file name, such as chain-menu-2026-09-29.pdf.
• Their Google Business Profile: stated hours, categories, star rating, number of reviews, and the 30-50 most recent reviews copied as text (sort by newest).
• Their last dozen social posts: what they push tells you what they want to sell more of.
• Their current ads: the Meta Ad Library shows ads running now across Facebook and Instagram and needs no login; the Google Ads Transparency Center shows ads on Search, YouTube and Maps. Search each rival's page name or web address.
• One old snapshot: the Internet Archive's Wayback Machine often has a copy of a price page from a year ago. The difference between then and now is a finding in itself.

Put each rival's files in one folder. If you use Gemini Notebook (formerly NotebookLM), or a Project in ChatGPT or Claude, upload the folder there so the assistant answers from those documents and cites them.

The ad libraries are usually the surprise. An illustrative dog groomer checking a rival found the same "first groom half price" ad had been running for five months. That tells you two things: the offer probably brings in enough first visits to keep paying for, and the rival relies on it. Either could shape your own reply.

Block 2: profile each rival against one fixed template (45 minutes)

Use the same prompt for every competitor so the profiles line up later. The instruction to use only your sources, and to write "not found" instead of guessing, is what keeps this honest.

You are helping me research a competitor of my business.
My business: [one line: what I sell, to whom, at what price level].
Competitor: [name].
Use ONLY the sources I have attached or pasted below. For every line,
give the source and its date in brackets. If the sources don't say,
write "not found". Do not guess and do not search elsewhere.

1. What they sell: top 5 items or services, with prices exactly as shown
2. Who they appear to target, with the evidence
3. Opening hours, availability or delivery area as stated
4. Their main promise in their own words (quote the headline)
5. Offers or promotions running now, including ads from the ad libraries
6. What customers praise (from the reviews provided)
7. What customers complain about (from the reviews provided)
8. What changed compared with the older snapshot
9. Three things they appear to do better than my business, then three worse

An excerpt of what came back for the chain (illustrative):

1. Flat white $4.10; oat milk add $0.60; breakfast wrap $6.95; pastry and coffee deal $6.50 [menu PDF, 29 Sep]. 2. Targets young professionals and commuters [app offer, ad library]. 3. 6:30am to 7pm weekdays, 8am to 6pm weekends [Business Profile, 29 Sep]. 5. "Opening week: any drink $1 in the app" [Meta Ad Library, ad started 22 Sep]. 7. Not found: no reviews yet.

Two fixes before this goes anywhere. Line 2 is a guess dressed up as a finding: an app discount doesn't prove who the shop targets, so either delete it or label it "my guess". And the menu listed a seasonal drink the owner couldn't find in the PDF; the assistant had pulled it from a general web result about the chain despite the instruction. Deleting unsupported lines takes two minutes and saves you acting on fiction.

Keeping the assistant inside your sources matters most with chains and common names. An illustrative florist found her profile of a rival quoted a 4.1-star rating when the real figure was 4.7: the assistant had blended in reviews of a florist with the same name in another town. The source-and-date brackets are how you catch that; any line without one gets checked.

Block 3: mine rivals' reviews for complaints you can fix (45 minutes)

Reviews are the richest free data you'll find, because customers describe exactly what disappointed them. Paste each rival's recent reviews, then your own, and ask for themes with evidence:

Below are [N] recent reviews of [competitor], numbered.
Group the complaints and the praise into themes. For each theme give:
a short label, how many reviews mention it, the review numbers you
counted, and two exact quotes. Count each review once per theme.
Then list which complaint themes also appear in MY reviews (second list).

The result for the brunch café, trimmed:

Asking for the review numbers is what lets you audit the counts. Spot-check two themes by reading the numbered reviews yourself. Assistants tend to overcount and to merge neighbouring ideas: in one illustrative run for a bakery, "expensive" came back with nine mentions, but four of those reviews actually said "pricey but worth it", which is praise with a caveat, not a complaint.

For the café owner, the finding is plain. The brunch rival's customers love the food and hate the Saturday wait, and her own reviews never mention waiting. That's a gap she can fill. The chain has no reviews yet, so she reads a sample of reviews from two of its other branches and treats them as a hint about what to expect rather than as evidence about this site. If you want to keep listening after the afternoon ends, AI social listening covers the ongoing version.

Block 4: put everyone in one comparison grid (40 minutes)

Paste the four profiles and the review themes into one conversation and ask for a table: rows are the factors customers choose on, columns are the rivals plus you, and any cell without a source says "not found". Then correct it by hand. The café's grid, after corrections:

Read the grid two ways. First, find the rows where everyone looks the same, because nobody wins there and competing on them costs money. Second, find the rows where one rival is uniquely weak or you are uniquely strong. Here, the café is the only place with quiet seating and good coffee before 8am apart from the chain, and it's the only option with space at weekends that isn't jammed.

Block 5: turn the gaps into two or three moves (30 minutes)

This is the block people skip, and it's the only one that changes anything. Ask the assistant to argue with you rather than cheer you on:

Act as a sceptical adviser to a small business owner.
Here is my comparison grid and the review themes.
Suggest five possible moves. For each: rough cost, time to set up,
which competitor it takes customers from, and what could go wrong.
Then list any assumption I seem to be making that the grid does NOT support.

Among its suggestions the assistant proposed matching the chain's $1 drinks, and flagged, fairly, that the owner was assuming commuters were her customers when the grid showed nothing about who actually visits before 9am. She checked her till data that evening: only 14% of weekday takings came before 9am. That one challenge changed the plan.

Her three moves, written as a table she pinned by the till:

A good rule for this block: don't set prices while a rival is running an opening promotion, because you'd be matching a number that expires in a fortnight. If pricing is the real question, read the limits of AI pricing research before you let a grid decide what you charge.

Where AI competitor research goes wrong

Most failures fall into a handful of patterns, and each has a quick tell:

• Old facts presented as current. A price from a cached listing or last year's menu. The tell is a missing or old date in the source bracket. Fix it by checking the live page.
• Merged identities. Branches of a chain, or businesses that share a name, blended into one profile. The tell is a rating or review count that doesn't match what you saw.
• Invented specifics when sources are thin. "They offer a loyalty scheme" with nothing behind it. Anything without a source gets deleted or tested.
• Flattery. Ask an assistant to compare a rival with your business and it tends to be kind to you. Asking for what the rival does better first, as the template does, counters that.
• False precision. "38% of reviews mention price" from a sample of 40 reviews with loose counting. Treat counts as rough ordering, not statistics.
• Blindness to anything offline. Queue length, how staff greet people, whether the toilets are clean. Twenty minutes buying a coffee at each rival fills that gap. The café owner's visit during the chain's opening week found a 12-minute queue at 8:15am, which no source pack would have shown.

If you plan to build customer profiles from what you learn, keep the same discipline: personas built from real data hold up far better than ones an assistant imagines. And for a wider habit of checking what research tools tell you, see how to check sources and citations in AI research.

Which assistant suits which block

Deep research modes, available in several assistants with allowances that vary by plan, are useful for background: how a chain runs its app, or what a mobile mechanic typically charges in general. For three to five nearby rivals, a normal conversation working from your own source pack is usually more accurate, because you control exactly what it reads.

On privacy, everything about competitors in this exercise is public. Your own figures are not. If you paste in takings or margins, do it on a business plan that doesn't train on your content, or switch off the model-training setting in the privacy settings of a consumer plan first.

Keeping the research fresh after the afternoon

Save the one-page output with the date in the name, and put a reminder in your calendar for three months' time. The second run is faster: you already have the template, the prompts and a list of source links, so refreshing the source pack and rerunning blocks 3 to 5 usually takes about 90 minutes.

A quick sum shows why that matters. The first afternoon costs about four hours. Four quarterly refreshes after that cost roughly six hours a year in total, which is less than one working day to know what four rivals are charging, promising and getting wrong. Between refreshes, a few pages are worth watching automatically; tracking competitors' prices and offers with AI covers how to do that without being flooded with alerts.

To check the afternoon earned its keep, look back after eight weeks and ask two questions. Did the moves actually happen, with the owners and dates you set? And when you spot-check five facts from the grid against the live sources, are they still right? The café owner's answers were yes and four out of five: the chain had already ended its $1 offer, exactly as expected, and her Saturday pre-orders had reached 19 a week.

Competitor research with AI: follow-up questions

Is it acceptable to research competitors this way?

Reading what competitors publish is ordinary market research: their website, public reviews, social posts and the ads shown in public ad libraries. Stay away from posing as a customer to extract confidential details, scraping sites at scale against their terms, logging in with fake accounts, or contacting a rival to discuss prices. If a particular method worries you, ask a solicitor before relying on it.

How often should I repeat the afternoon?

Once a quarter suits most small businesses, plus whenever a new rival opens or an existing one changes something big, such as prices, opening hours or delivery. The second run is quicker because the prompts and template already exist; most owners get it done in about 90 minutes by refreshing the source pack and rerunning the review and grid steps.

Should I paste my own sales figures into the AI?

You rarely need to. The comparison works with public facts plus a one-line description of your business. If you do want the assistant to weigh margins or sales, use a business plan that doesn't train on your content, or switch off the model-training setting in a consumer plan's privacy settings first, and share only the figures the question needs.

Further reads

• Perplexity vs ChatGPT for Business Research — Which research assistant cites sources better for this kind of job.
• Gemini Notebook (Formerly NotebookLM) for Small Business Teams — Set up a source-grounded notebook for your competitor pack.
• How to Test a New Service Idea With AI Before You Launch It — Test the gap you found before building a new offer around it.
• How to Do Keyword Research With AI for a Small Business Website — See which searches your rivals win and which are still open.
• How to Catch Outdated Information in AI Answers — Spot stale prices and hours before they reach your grid.
• Can a Small Online Shop Use AI to Compete With Bigger Brands? — How a small shop competes on more than price.
• How to Run a SWOT Analysis of Your Business With AI — Six stages and copyable prompts for a SWOT built on your own figures and reviews, with a podiatry clinic example and a TOWS step that turns it into actions.
• How to Write a Business Plan With AI, and What to Check Yourself — Draft each section of a business plan with AI from your own facts, build the financials yourself, and check the claims lenders and partners will test.
• How Consultants Use AI for Client Research Before Discovery Calls — A research routine for discovery calls: scale the prep to the deal, make AI cite everything, check what matters and turn findings into sharper questions.
• How to Find B2B Prospects With AI Research Tools — A seven-step method for turning your best customers into a checked list of look-alike companies, named buyers and timely reasons to get in touch.
• How to Write a Sales Deck With AI That Wins Meetings — Research one buyer, draft the storyline in plain text, turn every headline into a claim, and only then let an AI slide tool do the design.
• How Small Businesses Actually Use AI in Marketing: 15 Examples — Fifteen specific marketing jobs small businesses hand to AI, each with the steps, a sample, the time it takes and the catch to watch for.
• What AI Can and Can't Do for Your Local SEO — Which local SEO jobs AI does well, which it can't touch, and an afternoon's worked example for a dry cleaner, mapped against Google's own ranking factors.
• Are AI-Generated Customer Personas Accurate? How to Check — Why AI personas read convincingly and still miss, plus a two-hour audit with source tagging, data checks, recognition calls and prompts that show their working.
• Can AI Write a Marketing Plan for a Small Business? — The input pack AI needs, a three-prompt sequence, how to red-pen the first draft, and a veterinary practice's 90-day plan with budget and owners.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: Meta Ad Library and Google Ads Transparency Center public search pages; Google help on Gemini Notebook (formerly NotebookLM); vendor help pages for ChatGPT and Claude Projects.