An AI Competitor Research Template That Keeps Unknowns Visible

As of 2026-08-16, no verified performance, cost, user, or revenue data supports the competitor claims in this article. That absence is the point: beginner research should preserve unknowns instead of letting an AI system quietly convert them into facts. Start with one public source per competitor. Record each claim, its source date, the comparison field it might support, and what a person still needs to confirm. Do not pass the result into strategy, sales copy, or product planning until those blanks are resolved.

The three-line answer:

1. Capture claims from one public source without rewriting them as conclusions.
2. Separate quoted evidence, your inference, and every missing comparison field.
3. Require a named human check before the research moves to the next task.

The polished summary is where trouble begins

A beginner often asks for a competitor comparison and receives a confident table. The rows look complete. The language sounds neutral. Missing information has vanished.

But a neat table can conceal several different evidence problems. A feature may appear in marketing copy but not in the product. A price may exclude required extras. A page may have changed since someone last checked it. Two competitors may describe similar capabilities with different boundaries. An absent claim may mean “not offered,” or merely “not found on this page.”

The safe first deliverable is therefore not a competitive analysis. It is a source map.

A source map records what a public page actually supports. It also exposes what remains unanswered. Its job is to slow down interpretation just enough for a person to catch unsupported leaps.

A blank cell is more useful than a plausible answer with no traceable source.

For beginners, one source at a time is a useful constraint. It keeps the evidence trail readable and makes corrections cheap. Broader research can follow after the first source map reveals which questions deserve another source.

The template separates three kinds of statements

Use three labels throughout the document:

• Observed: The source explicitly states or visibly shows it.
• Inferred: The researcher interprets the observed material.
• Unknown: The available source does not settle the question.

These labels prevent a common collapse. Consider a fictional “convenience store BOGO deals app.” Its public page says users can browse nearby offers. That supports an observed claim about browsing offers. It does not prove complete store coverage, live inventory accuracy, or better savings than another app.

Those extra conclusions may sound reasonable. They are still inferences or unknowns.

Here is the source-map header:

Field	Entry
Competitor label	Use a neutral placeholder during drafting
Public source title	Exact page or document title
Source URL	Direct link to the evidence
Published or updated date	Date shown by the source, or “not shown”
Date checked	The date a person opened the source
Source type	Product page, help page, policy, pricing page, or public document
Research question	One question this source may answer
Reviewer	Person responsible for the final check
Status	Draft, needs review, verified, or rejected

If a source has no visible publication or update date, write “not shown.” Do not substitute the access date. They describe different things.

[Artifact caption: A real source-map screen should show the public page beside the completed evidence fields, including the visible “not shown” and “unknown” entries.]

Claims need smaller containers

Do not paste an entire page into a summary box. Break it into individual claims that can be checked independently.

Claim ID	Source wording or faithful note	Location	Status	Comparison field	Human check
C-01	What the source explicitly claims	Section heading or page area	Observed	Core function	Confirm current availability
C-02	A reasonable interpretation	Evidence used for inference	Inferred	Target user	Check another direct source
C-03	Information not present	Page searched	Unknown	Price conditions	Find an official pricing source

Keep source notes close to the original meaning. Quotation marks should be reserved for exact wording. A paraphrase should be marked as a paraphrase.

“Location” can be a heading, table row, FAQ item, or visible screen region. The goal is not academic citation style. The goal is letting another person find the evidence without repeating the entire search.

If another person cannot relocate the evidence, the claim is not ready to travel.

Comparison fields should expose missing symmetry

Competitor tables become misleading when one company has detailed documentation and another has a thin public page. The first appears stronger because more information was available, not necessarily because its product was better.

Create the comparison fields before interpreting the sources:

Comparison field	Competitor A	Competitor B	What must be confirmed
Intended user	Unknown	Unknown	Explicit audience statement
Core job	Observed claim	Observed claim	Whether the jobs are equivalent
Included capability	Unknown	Unknown	Current product availability
Important exclusion	Unknown	Unknown	Documented limitation
Price basis	Unknown	Unknown	Billing unit and required extras
Data handling	Unknown	Unknown	Current public policy
Support boundary	Unknown	Unknown	Included channel and conditions
Evidence date	Not shown	Not shown	Publication or update date

This matrix does not need to look impressive. It needs to reveal asymmetry. Never turn “unknown” into “no.” Never score a field merely because one source describes it more clearly.

A score can wait until the evidence is comparable and the decision criteria are explicit. Many beginner projects do not need one at all.

The handoff gate catches premature conclusions

Before research moves into positioning, content, outreach, or a buying decision, a person should review every row.

Use this gate:

• Each claim points to a direct public source.
• The source date is recorded, or marked “not shown.”
• The date checked is recorded separately.
• Observations and inferences have different labels.
• Missing information remains “unknown.”
• Similar claims have been checked for equivalent scope.
• Price entries include their billing basis and conditions.
• Negative claims are supported, not inferred from silence.
• The reviewer has reopened the source.
• Rejected claims remain in the record with a reason.
• The final comparison states its limits.
• The next task receives only reviewed rows.

The reviewer is not there to make the document sound smoother. The reviewer decides whether each claim is supported, needs qualification, or should be removed.

For higher-impact decisions, add a “decision relevance” column. A missing detail about typography may not block a workflow choice. A missing detail about data handling might.

Failure modes belong inside the artifact

This template cannot prove that a public claim is accurate. It only proves that the claim appeared in a specific source when checked. Public pages may be incomplete, outdated, ambiguous, or promotional.

One source is also insufficient for a broad verdict. The constraint is useful for building the first clean evidence unit, not for declaring a market winner.

Other failure modes include comparing different product tiers, treating a search snippet as the source, recording an undated screenshot without context, and asking an AI system to fill blanks from general knowledge. Each shortcut makes the output look more complete while weakening its audit trail.

There are no verified outcomes here showing that this method improves accuracy or decisions. There is also no verified duration or cost estimate. The recommendation is procedural: visible uncertainty is easier for a human reviewer to challenge than hidden uncertainty.

The template is complete when the uncertainty is legible, not when every cell contains an answer.

The final decision is a stop rule

Use the source map before any competitor conclusion. Keep observed evidence, inference, and unknowns separate. Then stop the handoff if a decision-critical claim lacks a direct source, a source date status, or human confirmation.

The primary action is simple: copy the source-map fields and complete them for one public competitor page before creating a comparison.

Related build logs

• AI Research Workflow for Beginners: One Bounded Question, One Verifiable Ledger
• AI Workflow for Beginners: A Five-Box Map Before Automation