AI competitor analysis without invented numbers

AI competitor analysis works well when the job is to gather public information, compare what companies currently offer and keep a record of what changed. An agent can read pricing pages, product documentation, help articles and other public pages far faster than a person checking each source by hand. The hard part is deciding what belongs in the final memo. A competitor's price, feature, plan limit or launch date is a fact only if the agent can point to evidence for it.

That distinction matters because competitor research often ends with a decision. Someone may use the memo to change a price, rewrite a product page or decide which feature to build next. A plausible number in the wrong cell can travel a long way before anyone notices. We prefer a research process where the agent does the collection and comparison, while a person remains responsible for claims that will affect the business.

Give the agent a research brief, not a blank search box

A useful brief says exactly what the agent is trying to establish. Instead of asking for a general competitor analysis, specify the companies, pages, questions and date of the research. For example: compare the current plans of four named competitors, record the monthly price shown on each official pricing page, list the features included in each plan, and flag anything that cannot be verified.

This also gives the agent a stopping point. If a company does not publish a price, the correct result is not an estimate based on an old article or a guess based on a nearby plan. It is 'not publicly listed' with the source checked. The same rule applies to customer counts, revenue, usage limits, release dates and claims about what a product can do.

• Name the competitors and the exact questions the research must answer.
• Set a source preference, such as official pricing and product pages before third party coverage.
• Give the research an as-of date so old information is not quietly mixed with current information.
• Tell the agent to mark missing or conflicting evidence instead of filling gaps.
• Define the final deliverable, such as a comparison table plus a short list of changes worth checking.

Separate facts from comparisons

A competitor memo becomes much easier to trust when it separates what the source says from what the researcher thinks it means. 'The Pro plan lists 20 seats' is a factual claim that should have a source. 'This makes the plan more suitable for larger teams' is an interpretation. Both can be useful, but they should not look like the same kind of evidence.

The same separation helps with calculations. An agent can compare two published prices or count features across published plans. It should not turn an unavailable figure into a precise estimate just because the memo looks incomplete without one. Precision is not evidence.

Make source checking part of the work

Competitor pages change. A pricing page can be updated without a press release, and a feature can move between plans while old articles remain searchable. An agent doing this work should capture the page it relied on, the relevant wording or data point, and the date it checked the page.

Conflicts need a visible rule too. Suppose an official product page lists a feature while an older comparison article says the feature is unavailable. The memo should not quietly choose whichever source produces the cleaner table. The current official page may be the better source, but the conflict can still be recorded so a reviewer knows why the result changed.

1. Open the preferred source and collect the requested fact.
2. Record the source and date beside the fact.
3. Check another relevant source when the claim is unusually important or the pages disagree.
4. Mark the claim as verified, conflicting or unavailable.
5. Only then use the information in a comparison or recommendation.

A missing number is a valid research result. 'Not published on the checked page' tells the decision maker more than an unsupported estimate does.

Ask the agent to flag changes, not invent explanations

One of the more useful jobs in competitor analysis is change detection. An agent can revisit the same public pages and compare the current information with an earlier research snapshot. It can flag that a plan now contains a feature, that a listed limit has changed or that a pricing page has been reorganized.

The agent should stop at the observation. If a feature disappeared, that does not prove the company discontinued it. If a price increased, that does not prove the company is responding to a rival. Those explanations require evidence. A good memo says what changed and leaves the reason open unless a source establishes it.

This is where a research analyst brief can save time. Delagents has a research analyst template for delegating structured research while keeping the expected deliverable and review point explicit.

Put a person at the decision boundary

The agent can gather pages and produce a draft comparison. The person signing off should decide which findings are reliable enough to use. That review does not need to mean rereading every page. It can focus on the claims that would change a decision: prices, plan limits, product capabilities, major launches and any recommendation that follows from them.

A useful approval question is simple: 'Could I open the cited source and see why this number or claim appears here?' If the answer is no, the finding goes back for research or gets removed. That gives the agent room to do the repetitive work without giving it permission to manufacture certainty.

For work that can trigger an external action, the boundary should be even clearer. Research can inform a pricing change, product announcement or sales message, but the research itself should not silently turn into a published statement. A person should see the proposed action before it leaves the workspace.

A practical memo format

The final document does not need to be long. We would rather have a short table with traceable facts than several pages of confident prose. A useful structure is: scope and research date, competitor comparison, changes since the previous check, unresolved conflicts, and observations that need human judgement.

For each important fact, keep the evidence close enough that a reviewer can verify it without starting the research again. If the agent cannot establish a claim, say so. If two sources disagree, show the disagreement. If the agent makes an interpretation, label it as one.

Start with one narrow competitor question this week. Pick three or four public pages, define the facts you want collected, require a source and check date for every factual claim, and tell the agent to leave unknowns blank. Then review the finished memo before anyone acts on it. That small test will show whether your research brief gives the agent enough room to work and enough boundaries to stay honest.