Writing

• Marketing Workflows
• 4 October 2026

AI tools for competitor analysis: a workflow and prompts that work

The AI tools that help most with competitor analysis fall into three groups: general assistants with a research mode (ChatGPT deep research, Claude, Gemini, Perplexity) for reading and summarising, specialist data tools (Semrush, Similarweb, the Meta and Google ad libraries) for evidence you can’t get by asking a chatbot, and competitive intelligence platforms (Crayon, Klue) for teams that need it monitored continuously. None of them does the analysis for you. The useful part is a workflow where you collect the evidence and the AI sorts it.

Adoption has moved fast. Crayon, a US competitive intelligence software company, says in its 2026 State of Competitive Intelligence report that 80% of the CI, sales and revenue leaders it surveyed use AI to generate sales-facing competitive content, up from 61% in 2025 and 25% in 2024. Crayon sells software in this space and doesn’t publish the 2026 sample size, so read it as a direction of travel rather than a census.

What AI tools help with competitor analysis in marketing strategy?

I’d group them by what they are good for, because the most common mistake is asking a general chatbot to do a job that needs a data tool.

General assistants with a research mode

OpenAI describes deep research in ChatGPT as a way to “plan, research, and synthesize complex questions into a documented report”, and says every output includes citations and a sources section so you can check it. Claude and Gemini offer similar research features. They are very good at the dull part of competitor analysis, which is reading 30 product pages, pricing pages and press releases and putting the facts into one table. They are poor at telling you anything that isn’t published, and they won’t say so unless you ask.

Specialist data tools

Semrush’s Keyword Gap compares up to five domains and sorts the results into groups such as “Missing” (every competitor ranks and you don’t) and “Untapped” (at least one competitor ranks and you don’t). That second list is where most content plans should start, because it shows demand a rival has already proved exists.

Similarweb estimates traffic from four sources: websites that share their own analytics, a “contributory network” of consumer products that collect anonymous device behaviour, data partnerships, and public data it extracts. Its methodology page explains the models. It’s useful for trends and channel mix, and I wouldn’t put its exact visit numbers in front of a board, especially for a niche UK site.

The ad libraries are free and underused. Meta’s Ad Library lets anyone search all active ads running across Meta’s products. Ads that deliver in the EU stay visible for a year after their last impression, with extra detail, but a UK-only commercial campaign shows only while it is live, so if you want a record you have to capture it as you go. Google’s Ads Transparency Center lets you search by advertiser or website and filter by region and date.

Competitive intelligence platforms

Crayon monitors competitors’ news, social media, PR and announcements and turns them into alerts, battlecards and newsletters for sales teams. Klue pulls from CRM data, recorded sales calls, win-loss interviews, internal documents and the public web, and adds a human verification layer. Both make sense when you have a sales team losing deals to named rivals every week. If you’re a brand manager with three competitors and a quarterly plan, a spreadsheet and a disciplined monthly hour will get you most of the way.

Why can’t you just ask ChatGPT about your competitors?

Because the answer will be fluent and partly wrong, and you won’t know which part. The Tow Center for Digital Journalism at Columbia tested eight AI search tools with 1,600 queries in a March 2025 study by Klaudia Jaźwińska and Aisvarya Chandrasekar. Each query gave the tool a direct excerpt from a news article and asked it to name the headline, publisher, date and URL. Collectively the tools got more than 60% wrong. Perplexity was wrong on 37% and Grok 3 on 94%, and the premium versions were more likely to give “definitive, but wrong” answers than to admit they didn’t know.

That was a test of finding a known article, which is easier than describing a competitor’s strategy. So when we ask a chatbot “what is Oatly’s positioning in the UK?” and paste the answer into a strategy deck, we’re presenting a confident summary of whatever the tool happened to read, possibly a three-year-old trade press interview, as if it were research. The fix is to reverse the order: you gather the primary material, and the AI’s job is to read and sort it.

A workflow for AI competitor analysis

This is the process I’d run for a brand or campaign plan. It takes a day the first time and an hour a month after that.

1. Write down the decision. “Should we launch a cheaper tier?” needs pricing pages and reviews; “Why are we losing search traffic?” needs Keyword Gap and content audits. A competitor analysis with no decision attached tends to become a 40-slide deck of logos.
2. Pick three to five competitors, including one you’d rather not think about. Usually that’s the cheap substitute or the new entrant customers mention, and rarely the rival the leadership team enjoys arguing about.
3. Collect the evidence yourself. For each competitor, save the homepage, pricing page, two or three product pages, the last six months of press releases, current ads from the ad libraries, a Keyword Gap export and a sample of recent customer reviews. Save them as files or text, with the URL and date.
4. Have the AI extract into a fixed grid. Same columns for every competitor: target customer, main promise, proof points, price points, channels, recent changes, and complaints in reviews. A fixed grid stops the model writing a nicer essay about the competitor with the nicer website.
5. Compare and look for gaps. Ask where competitors all say the same thing, where nobody is talking to a customer group, and where reviews contradict the marketing.
6. Verify before anything leaves your laptop. Every number, price and claim in the output gets checked against the saved source. Anything you can’t trace gets deleted.
7. Turn it into a decision and a watch list. One page: what we’ll do differently, and the three signals that would make us rethink. Then set the monthly check.

Steps 3 and 6 are the ones we skip, because they’re boring, and they’re the ones that make the rest trustworthy. If you’ve already built a broader process, this slots into the research stage of an AI marketing workflow.

Prompts for competitor analysis that work

These assume you’ve uploaded or pasted the evidence from step 3. The instruction to use only the supplied material matters more than any clever wording.

Extraction into a grid:

You are helping with a competitor analysis for [our brand], a [category] brand selling to [customer] in the UK.
Using ONLY the documents I've attached, fill this table for each competitor:
target customer | main promise | proof points | prices (with date seen) | channels in use | changes in the last 6 months | top 3 complaints in reviews.
For every cell, add the document name it came from. If the documents don't say, write "not found". Do not use outside knowledge.

Positioning gaps:

Using the completed table, list:
1. Claims that three or more competitors make in near-identical words.
2. Customer groups or needs that none of them address directly.
3. Places where customer reviews contradict what the brand claims.
For each point, quote the evidence and name the source document.

Ad library review (paste screenshots or ad text with dates):

Here are the ads [competitor] is currently running on Meta and Google, with the date I captured each.
Group them by message and by audience they appear aimed at. Which messages appear most often? What has changed compared with the set I captured on [previous date]? Don't speculate about spend or performance; the libraries don't show it.

Search gap (paste a Keyword Gap export):

This is a Semrush Keyword Gap export showing "Untapped" keywords for [our domain] against [competitors].
Cluster the keywords into topics, estimate which topics show buying intent versus research intent, and suggest the five topics where a single strong page could compete. Flag any cluster where the competitor's ranking page looks like a product page rather than an article.

The challenge prompt, which I’d always run last:

Act as a sceptical strategy director reviewing this analysis. Which conclusions rest on a single source? Which might be out of date? What would we need to see to believe the opposite?

What competitor data do UK marketers have that AI tools miss?

Several of the best UK sources are free, public and rarely read by AI tools unless you hand them over.

Companies House filings

The Companies House register gives free access to filing histories, accounts, officers and charges for every UK company. There’s a catch for analysing smaller rivals. According to gov.uk guidance, a company counts as small if it meets two of: turnover of £15 million or less, £7.5 million or less on its balance sheet, and 50 employees or fewer, and small companies can choose not to file their profit and loss account. Micro-entities file only a simplified balance sheet. So a challenger brand’s accounts may tell you its net assets and nothing about its revenue, and if an AI tool confidently quotes a private UK company’s turnover, ask it which filing that came from.

Officer appointments, new charges against assets and changes of registered office are still useful signals: a new finance director with a private equity background often hints that something is about to happen.

ASA rulings

The Advertising Standards Authority publishes its rulings every Wednesday, searchable by advertiser. A rival’s upheld ruling tells you which claims they’ve had to drop, and an informally resolved complaint tells you someone has been watching them closely. It’s also the cheapest way to learn where the line is before your own team writes the same claim.

Customer reviews and your own sales team

Reviews on Trustpilot, Amazon, app stores or G2 are the best evidence of what customers actually experience, and AI is good at clustering a few hundred of them into themes. Your own sales or customer service team is the other underused source: the objections they hear about competitors are primary research that no tool can scrape. Klue’s whole pitch rests on combining exactly this kind of internal material with public data.

How do you keep competitor research out of other people’s models?

The internal material is the sensitive part. Win-loss notes, CRM exports and pricing discussions shouldn’t go into a personal chatbot account. OpenAI’s enterprise privacy page says it does not train its models on data from ChatGPT Business, Enterprise or the API by default; consumer accounts work differently, and settings vary. Check which plan your team actually uses before anyone uploads last quarter’s lost-deal analysis, and keep a short rule in your AI guidelines about what can and can’t be pasted in. We spend months protecting a pricing strategy and then share it with a free chatbot account to make the slides look tidier.

On the public side, stick to material anyone can see. Reading a competitor’s website and ad library is research; using a fake identity to sit through their sales demo is something your legal team would want to hear about before, not after.

How often should you run competitor analysis?

For most UK marketing teams, a full refresh once or twice a year, before planning, and a monthly hour of monitoring is enough. The monthly check is small: new ads in the libraries, pricing page changes, new filings at Companies House, ASA rulings, and any shift in the Keyword Gap. A page-change monitor or the Wayback Machine helps with pricing pages, which competitors change without a press release.

Continuous platforms such as Crayon or Klue pay for themselves when sales teams face the same rivals in live deals and need answers that week. Crayon’s 2026 report claims that teams making competitive intelligence available through a chatbot report revenue impact at 79%, against 43% without; that’s a vendor’s survey of its own field, so treat it as a hypothesis worth testing, not a business case, and measure it the way you would any other tool (the method in measuring the ROI of AI in marketing works here).

If you want to see how this kind of structured research feeds targeting, the approach in AI customer segmentation uses the same discipline of collecting evidence first. At ACTAsia, where I lead AI adoption in marketing, the rule I’d apply is the same one: the model sorts, people source and check.

A good first step this week: pick the two competitors you mention most in meetings, capture their live ads from the Meta Ad Library and Google’s Ads Transparency Center today, and save them with the date. Next month, run the same capture and ask an AI tool what changed. Which of your competitors would surprise you most?

FAQs

Can ChatGPT do a competitor analysis?

It can do part of one. ChatGPT’s deep research mode can read many public pages and produce a cited report, which saves hours of reading. It can’t see anything unpublished, such as private pricing, campaign results or a small UK company’s revenue, and AI search tools often misattribute sources. Use it to sort evidence you’ve collected, then check every figure against the original page.

What is the best free tool for competitor analysis?

For UK marketers, the Meta Ad Library and Google’s Ads Transparency Center are the most useful free tools, because they show the ads competitors are running now. Companies House adds free filing histories, officers and accounts, and ASA rulings show which competitor claims have been challenged. A general AI assistant can then summarise what you collect from these sources.

How do I see a competitor’s ads in the UK?

Search the competitor’s name in the Meta Ad Library to see all their active ads across Facebook, Instagram and Meta’s other products, and in Google’s Ads Transparency Center, where you can filter by region and date. Meta keeps EU-delivered ads for a year after they stop, but UK-only commercial ads show only while live, so capture them with the date.

Is it legal to use AI for competitor research?

Analysing publicly available information, such as websites, ads, filings and published reviews, is normal market research. The risks are elsewhere: pasting confidential internal data into an AI tool that may use it for training, breaching a site’s terms by scraping it, or misrepresenting who you are to get competitor information. Check your AI tool’s data settings and your company’s guidelines first.

How accurate are AI competitor analysis tools?

It depends on the tool. Data tools like Semrush and Similarweb produce modelled estimates that are better for trends than exact figures. AI assistants can misread or misattribute: a Tow Center study at Columbia found eight AI search tools gave incorrect answers to more than 60% of queries when asked to identify the source of news excerpts. Always verify against the original.

Sources

1. Crayon, The 2026 State of Competitive Intelligence, 2026.
2. OpenAI Help Center, Deep research FAQ, accessed October 2026.
3. Semrush Knowledge Base, How to use Keyword Gap?, accessed October 2026.
4. Similarweb, Similarweb Data Methodology, accessed October 2026.
5. Meta Transparency Centre, Meta Ad Library tools, updated 24 August 2023.
6. Google My Ad Center Help, About advertiser verification, accessed October 2026.
7. Crayon, Crayon competitive intelligence platform, accessed October 2026.
8. Klue, Official Information About Klue, accessed October 2026.
9. Columbia Journalism Review, Tow Center (Klaudia Jaźwińska and Aisvarya Chandrasekar), AI Search Has a Citation Problem, 6 March 2025.
10. Companies House, Find and update company information, accessed October 2026.
11. GOV.UK, Prepare annual accounts for a private company: Micro-entities, small and dormant companies, accessed October 2026.
12. Advertising Standards Authority, Rulings, accessed October 2026.
13. OpenAI, Enterprise privacy at OpenAI, accessed October 2026.