How to Use AI to Compare B2B Competitor Messaging

Use AI as an extraction and comparison assistant, not as a mind reader. Give it the captured text from two comparable public pages and ask it to identify only what each page explicitly says about its audience, problem, offer and visible proof. Require a short supporting passage and a location for every observation, and require NOT STATED when the evidence is missing.

Then check every row yourself and separate observation, interpretation and missing evidence. A public page can show what a company chose to publish on a particular date. It cannot reveal private strategy, prove product quality or tell you how customers responded.

What can a public messaging comparison tell you?

A public page is direct evidence of the message published there when you checked it. You can examine its named audience, stated problem, offer and visible proof. You cannot assume that a headline represents everything the business believes, that a missing problem has been overlooked or that a customer logo proves suitability.

The House of Commons Library's source-evaluation guidance recommends considering accuracy, authority, relevance, transparency and purpose. A vendor page is authoritative for its visible wording but has a promotional purpose. It is not independent confirmation of every claim.

Start with comparable pages and a dated capture

Choose pages that perform the same job, such as two product overviews, service pages or homepages. A detailed case study and a short homepage give each business a different opportunity to explain itself.

For each page, record:

• company and page title;
• full URL;
• page type;
• date checked;
• the exact page text supplied to the AI tool.

Keep the captured text with your notes. If a page is dynamic, narrow the exercise to passages you can verify. Follow your organisation's policy and the site's terms when collecting material.

Compare four messaging dimensions

Use the same questions for both sources.

These dimensions are a practical lens, not a universal formula. They keep the extraction and review consistent.

Give AI a restriction-based comparison instruction

Supply the two page captures in clearly labelled source sections. Then use an instruction such as this:

Task: Compare Source A and Source B using only the supplied text. For audience signal, problem signal, offer and visible proof, return: source; short supporting passage; page location; factual observation; cautious interpretation; missing evidence. Rules: If the source does not support a field, write NOT STATED. Do not infer private strategy, priority audience, product quality, customer opinion, market position or performance. Do not choose a winner. Flag any part of the comparison that is not like for like.

This wording cannot guarantee accuracy. It makes the expected evidence and forbidden inferences visible for review.

Worked example: two current platform pages

This example compares the Asana product overview with the monday.com homepage. Both are top-level platform marketing pages, although their templates differ. The visible pages were checked on 9 September 2026 and may since have changed.

The headline contrast is useful when kept small. Asana leads with “The only work management platform built for scale”. monday.com leads with “People and agents working as one team”.

At the checked date, Asana's headline puts more visible emphasis on scale and the work-management category, while monday.com's emphasises people and agents acting together. This does not show either company's complete positioning, strategy or success.

Check the AI comparison before using it

OpenAI's accuracy guidance says important information, quotations and references should be verified. Apply that principle even to a tidy output.

Use five checks:

1. Passage check: Can you find the supporting words on the original captured page?
2. Context check: Do the surrounding heading, image and paragraph preserve the meaning?
3. Page check: Are both sources serving a comparable purpose?
4. Inference check: Has an observation quietly turned into a claim about strategy, quality or performance?
5. Proof check: Is a proof signal clearly labelled as vendor-published, dated where possible and limited to what it actually shows?

Correct or remove any row that fails. A polished table with an invented passage is still wrong.

Turn the comparison into a question about your own message

Use competitor evidence to prompt a question you can investigate in your own business. It cannot replace customer conversations, sales evidence, product knowledge or performance data.

Possible questions include:

• Which reader should recognise themselves in our opening section?
• Which customer problem do we state explicitly before describing the offer?
• Can a reader tell what action or outcome our offer supports?
• Which proof can we substantiate, date and put in fair context?

Answer one question with your own approved evidence. If a competitor does not mention a problem on the reviewed page, do not call it a market gap. Record only that it was not stated there at that time.

If you need a broader method for organising research before making a content decision, read How to Use AI to Organise B2B Audience Research Without Losing Human Judgement.

Avoid four common shortcuts

• Do not ask AI to infer each competitor's strategy from a URL.
• Do not convert a missing statement into a weakness or opportunity.
• Do not compare vendor figures as though their definitions and dates match.
• Do not copy a competitor's distinctive language into your own marketing.

The value comes from a narrow, reviewable comparison. Stronger-sounding conclusions do not make weak evidence better.

Your next step

Choose two comparable public pages and complete one four-row table. Record the URLs and today's date, verify every passage, then finish with one question about your own messaging that needs your own evidence.

Further Reading

• How to evaluate sources, House of Commons Library
• Does ChatGPT tell the truth?, OpenAI Help Center
• Asana product overview
• monday.com homepage