TL;DR
- Use AI to structure and re-grade a competitor matrix, not to recall competitor features. Paste each competitor’s feature list pulled from their live site the same day; models have an April 2026-ish training cutoff and vendor pages change quarterly.
- A useful matrix has rows = 15 features, columns = your product + 3 direct competitors + 1 adjacent, cells =
strong / present / weak / missingon one shared scale. - Don’t grade on row count. Weight each feature by buyer importance (1-3) and sum the weighted score per product. The weighted total reveals the real picture.
- Anchor buyer importance in data, not opinion: pull the 50 most recent G2/Capterra reviews per product; a feature mentioned in 20%+ of reviews is high importance, 5-19% medium, under 5% noise.
- Best tool as of June 2026: GPT-5.5 for the same-day data pull (BrowseComp web-research 90.1%, ahead of Gemini 3.1 Pro 85.9% and Opus 4.7 79.3%); Claude Opus 4.7 for the brutally honest re-grade, since it handles uncertainty and pushes back harder on self-flattery.
The task
Your CEO asked for a competitive positioning doc by Friday. Sales has been losing 3 deals a week to competitor B and product is convinced it is a sales execution problem; sales is convinced it is a feature gap. You want a matrix that puts your product next to the three real competitors across the 15 features that actually drive a buying decision — and that tells you, with cold honesty, which of your “differentiators” are table stakes everyone has, and which two or three things you genuinely win on. Not a marketing one-pager. A tool the founder can read in 4 minutes and the head of product can act on Monday.
Where AI helps, and where it does not
AI is strong at two jobs: structuring the matrix, and the analytical second pass that sorts features into “table stakes” (everyone has it, no differentiation) versus “true differentiator” (only 1-2 players win, ranked by buyer importance). It is also useful for naming features your team over-claims on the website but actually under-delivers on.
Where AI fails: verifying competitor features. Shipped models carry a training cutoff (GPT-5.5, Opus 4.7, and Gemini 3.1 Pro all landed around April 2026), and competitor websites change quarterly. The feature page you remember is not the page live today. Pull each competitor’s feature list from their current site and docs the same day you build the matrix, and paste it in as raw text. Do not let the model recall from memory.
That cutoff problem is exactly why the tool you pick matters. As of June 2026, GPT-5.5 leads web-research accuracy on BrowseComp (90.1%, vs Gemini 3.1 Pro 85.9% and Claude Opus 4.7 79.3%), and it makes fewer tool calls per task, so it is the better choice for the same-day data pull and for a quick site: sweep of a competitor’s docs. For the re-grading pass, Claude Opus 4.7 is the better partner: it handles uncertainty more carefully and pushes back harder when your self-grading is flattering. You can run the whole thing in one model; splitting it just plays to each one’s strength.
A common failure mode: the model is too generous on your side and too cautious on theirs. You wrote “strong” in your feature list, the model agrees, and the matrix tells you what you wanted to hear. Tell it explicitly to challenge your self-grading, and ask for the 3 features where you rank weakest on the same matrix.
What to feed the AI
- Your full feature list with honest self-grading — strong / present / weak / missing — and a one-sentence justification per “strong” claim
- 3 direct competitors plus 1 adjacent (alternative tool buyers also consider)
- Each competitor’s feature list, pulled from their current website / docs the same day
- Your target buyer in one sentence: who, role, company size, what makes them switch
- The 5 features your buyer says they care most about (from sales calls and reviews, not your guess)
- The 5 deal-loss reasons sales has reported in the last 90 days
- The 3 features you have been selling against on the website that you suspect are weaker than you market
- Your honest 90-day roadmap on the gaps: what you will fix, what you will not
Score buyer importance from reviews, not opinions
The weakest link in most matrices is the “buyer importance” weight. If you guess it, the matrix just reflects whoever argued loudest in the room. Ground it in evidence instead.
Pull the 50 most recent reviews for your product and each direct competitor from G2 and Capterra, restricted to the last 12 months. Paste them in and ask the model to count which features get mentioned and at what rate. A practical threshold: features named in 20% or more of reviews are high importance (weight 3), 5-19% medium (weight 2), and under 5% is noise unless sentiment is unusually strong. Watch the direction too: a feature mentioned a lot in 1-star reviews is a churn risk, not a selling point. This gives every weight in your matrix a defensible source you can show the founder.
Copy-ready prompt
Build a competitor feature matrix.
My product features (with my self-grading): {paste with strong/present/weak per feature}
3 direct competitors + 1 adjacent (with feature lists pulled today): {paste each, named}
Target buyer (1 sentence): {description}
Top 5 buyer-importance features (from sales calls): {list}
Recent deal-loss reasons: {list 5}
Features I suspect we over-claim on the website: {list}
Return:
1) Feature matrix table — rows are features, columns are my product + the 4 competitors, cells are strong / present / weak / missing. Use the same scale across all products; do not grade me on a curve.
2) Table-stakes features — everyone has them, do not differentiate. Drop these from positioning.
3) True differentiator features — only 1-2 players win. Rank by buyer-importance score (1-3 each). The product with the highest buyer-importance-weighted score in this section wins on positioning.
4) "Stop pretending" list — features I claim or sell against where my actual capability is "present" or "weak." These need either a 90-day investment or a website rewrite.
5) "Honest narrative" paragraph — 100 words. Read like a memo to the founder, not a marketing deck.
6) One question I should ask sales this week to validate this matrix.
Challenge mode: name the 3 features where my product is actually the weakest in the matrix. Do not soften.Shorter variant — single-feature deep audit
Audit one feature of our product against the same competitors: {feature name}.
What we claim on the website: {paste}
What our actual capability is (be blunt): {paste honest assessment}
Competitor claims and capabilities (from their current website + docs): {paste each}
Return:
1) Truth-vs-marketing gap for each player, including us.
2) Buyer interpretation — what a buyer hears when they read our website on this feature.
3) The single sentence we should change on our website if we are over-claiming.
4) One question to ask a buyer in the next discovery call to test whether this gap is hurting us.Sample output
A few rows of a real matrix, with the buyer-importance weight in front so the weighted total does the talking:
| Feature (weight) | Us | Comp A | Comp B | Adjacent |
|---|---|---|---|---|
| Enterprise SSO (3) | weak | present | strong | missing |
| Time-to-first-value (3) | strong | present | weak | present |
| Pricing transparency (2) | strong | weak | weak | strong |
| API + webhooks (2) | present | strong | strong | weak |
| In-app reporting (1) | present | present | strong | weak |
Score strong = 2, present = 1, weak = 0, missing = 0, multiply by weight, and the picture changes from the raw checklist: you may lead on row count yet trail Comp B once SSO and API are weighted by what buyers actually decide on.
A useful “stop pretending” line: “Stop pretending we are competitive on enterprise SSO. Competitor B ships SAML, SCIM, audit logs, and SOC 2; we ship OAuth and a single ‘enterprise’ page screenshot. Either invest one quarter to close the gap or remove SSO from our enterprise pitch deck and reposition mid-market only.”
A useful honest narrative: “Across 15 features and the buyer-importance weighting, we genuinely win on 2: time-to-first-value (our onboarding clears in under 3 minutes vs Competitor B’s 27, A’s 12) and pricing transparency. We tie the field on 8 (table stakes). We lose on 5, three of which we currently market as wins. The product gap on enterprise SSO is the biggest contributor to lost deals over the last 90 days; closing it is a bigger lever than any ‘differentiator’ we could add this quarter.”
A useful single question for sales: “On the last 5 deals we lost, what did the buyer say after seeing Competitor B’s SSO page that we cannot match? I want the exact quote, not a paraphrase.”
How to refine
- If the matrix is just feature checklists: “For each feature, weight by buyer importance (1-3). Sum the weighted scores by product. The weighted total, not the row count, reveals the real picture.”
- If the model is too kind to you: “Re-score my product as if you were a hostile competitor at a sales bake-off. Name 3 features I will lose on in a head-to-head demo.”
- If ‘differentiator’ features are obviously table stakes: “A differentiator must be (a) buyer-important AND (b) won by only 1-2 players. Re-classify any feature that fails either test. The differentiator list usually shrinks; that is correct.”
- If the narrative reads like marketing: “Rewrite as a memo to a skeptical founder. Cut every adjective. Lead with what the matrix actually shows, not with how we feel about it.”
- If sales and product still disagree: “Add a ‘whose data wins’ column. For each contested feature, name what evidence (a real deal call, a competitor’s docs link, a customer ticket) would settle the disagreement.”
Common mistakes
- Self-grading too generously: pulls the whole matrix into “we win”; the matrix becomes a marketing artifact instead of a decision tool. Always include the “name our 3 weakest” challenge.
- Stale competitor data: features change quarterly; the page you remember is not the live page. Refresh competitor data the day you build the matrix.
- Treating all features as equally important: a 15-feature matrix where the buyer cares about 4 makes the other 11 a distraction. Weight by buyer importance.
- No “deal-loss reasons” input: the matrix becomes academic; deal-loss reasons are how you validate that the matrix matches the market.
- Confusing breadth with strength: having “a version of” a feature is not the same as winning it; if the competitor’s version is materially better, you do not “have” the feature in the buyer’s view.
- Publishing the matrix externally as-is: internal matrix is brutally honest; external comparison content needs simplification and selection. See the linked product comparison copy guide.
- Treating the matrix as static: features change, buyers change, competitive landscape changes. Rebuild quarterly, not annually.
- Ignoring the adjacent competitor: the buyer often considers a non-direct tool (Notion vs Linear, Figma vs PowerPoint); ignoring the adjacent option means your matrix misses the alternative the buyer might actually pick.
FAQ
- Should the matrix be shared internally only, or published externally?: Internal: brutally honest, all 15 features, all 4 competitors named. External: simplified, 5-7 features max, focused on the 2-3 dimensions where you genuinely win. See the product comparison copy article for the external version.
- How many competitors to include?: 3 direct + 1 adjacent. More than that and the signal dilutes; fewer and you miss the actual buying alternative.
- What if I do not know my competitor’s deeper capability beyond their website?: Set up trial accounts. The website is marketing; the product is the truth. For sales-led tools, request demos; many features that appear in the matrix as “strong” are gated demos that actually under-deliver. Mine third-party reviews too: scanning the last 12 months of a competitor’s G2 reviews surfaces the gaps their marketing hides.
- How do I handle a competitor with a totally different model (open source, enterprise-only)?: Add a “model fit” row above the feature matrix. A competitor that is 10x more powerful but enterprise-sales-only is not the same competitor for a self-serve SMB buyer; the model row tells you when to ignore them entirely.
- What if my honest matrix shows we should not be competing here?: That is a real finding. Either reposition to a segment where the matrix shifts in your favor, or commit to closing the gap with a clear 90-day plan. The matrix that confirms the team’s preferred narrative is the matrix that taught nothing.