Use ChatGPT to Build a Source-Linked Competitor Brief

A competitor brief gets risky when a clean paragraph hides where each claim came from. Use ChatGPT to structure the research, not to turn an unverified summary into a fact sheet.

Evidence label: documentation-based guide with fictional inputs and illustrative output. This is not a recorded ChatGPT research session and does not claim a measured research benefit.

What you need

• ChatGPT with the current web search or browsing capability available to your account. Names and availability vary by plan and surface.
• Three public competitor pages you are allowed to review.
• A question narrow enough to answer in one sitting.

OpenAI’s prompt engineering guidance recommends clear instructions, useful context, and iterative checking. OpenAI’s search guidance explains that search responses should still be checked against their linked sources.

1. Set a small fictional brief

Use this example:

Question: How do three project-management tools describe guest access for a small team?
Audience: an operations manager choosing what to investigate next
Date boundary: prefer pages updated or accessed in October 2026
Sources: only the official pricing, permissions, or help pages for AcmeBoard,
PlanNest, and TaskHarbor. If an official page is unavailable, mark the row unknown.
Do not compare prices. Do not recommend a winner.

The company names and pages are fictional. Replace them with public pages you can verify. Do not paste private customer or internal competitor material into an unapproved account.

2. Ask for a claim ledger first

Open a new chat and paste this prompt after the brief:

Research only the question and source boundary below.

Return a table with these columns:
competitor, exact claim, source page title, source URL, page date if shown,
source type (company statement or independent source), confidence (high,
medium, or unknown), and what the page does not establish.

Rules:
1. Use one row per material claim. Do not merge several claims into one sentence.
2. Quote or closely paraphrase only what the source supports.
3. Separate a company's claim from an independent fact.
4. If you cannot verify a claim from an allowed page, write unknown.
5. Do not infer a feature from a marketing phrase.
6. Do not recommend a product or invent a price.
7. Include the retrieval date and the exact URL for every row.

After the table, write:
A. three questions the human reviewer should check;
B. a short, neutral brief using only high-confidence rows;
C. a list of excluded claims and why they were excluded.

Keep the output as a claim ledger before asking for prose. A polished paragraph is harder to audit than a table with one source per row.

3. Build the review artifact

Copy the response into a document or spreadsheet with these columns:

Competitor	Claim	Source URL	Date shown	Company statement or independent source	Confidence	Reviewer note
AcmeBoard	guest access is described on the official permissions page	URL to the fictional page	date or unknown	company statement	high or unknown	confirm the wording

This is a template, not captured model output. The artifact passes its first check when every claim has a URL, every unknown remains visible, and a reader can tell what the company said versus what you inferred.

4. Review before sharing

1. Open every linked page and confirm that it contains the claim.
2. Check whether the page describes current behavior, a planned feature, or a general marketing promise.
3. Remove any row that relies on an unapproved source.
4. Keep the retrieval date beside claims that may change.
5. Ask a human owner whether the question is still the right one before writing a recommendation.

Do not treat a source link as proof that the conclusion is correct. It only makes the claim easier to inspect.

Troubleshooting

• The answer contains one citation for several claims: ask for one row and one source URL per material claim.
• A marketing phrase becomes a feature fact: quote the exact sentence and mark the capability unknown unless the documentation supports it.
• A source is missing or blocked: keep the row as unknown. Do not replace it with a search snippet or an invented detail.
• The brief recommends a winner: remove the recommendation and ask for a conditional comparison based on the stated question.

Limits and safe use

Public pages change. Store the retrieval date and inspect the pages again before a purchase, security review, or customer promise. Keep private strategy, credentials, personal data, and confidential competitor material out of the chat unless your organization’s policy permits that exact use.

TRY this on three public pages and a narrow question. SKIP using a fluent summary as evidence without opening its sources. USE the brief as a research handoff, not as an automatic buying decision.

Sources and next step

• OpenAI: Prompt engineering best practices for ChatGPT
• OpenAI: ChatGPT Search
• Use ChatGPT to turn product documentation into a safe customer reply draft
• Get The Weekly Verdict

Next, apply the same claim-ledger format to one internal research question with approved sources and a named reviewer.