TL;DR
A quarterly competitor refresh — 5 rivals, one consistent format — is a perfect AI task if you feed it real source material and run a fact-check pass. General-purpose models (ChatGPT, Claude, Gemini) are excellent at structuring pasted pages into a clean matrix and weak at recall: ask them to remember a competitor’s funding or ARR and they invent plausible numbers. The fix is mechanical. Paste sources, force a fact-check log that maps every claim to a source line, and flag anything unsourced [UNVERIFIED] before it reaches a deck. Below is the prompt, the recommended structure, and where dedicated tools (Crayon, Klue, Similarweb) earn their keep.
The task
You run a quarterly competitive refresh across five competitors. You want one consistent template covering positioning, pricing, go-to-market motion, strengths, and vulnerabilities — plus a “so what” section that turns the analysis into a decision (a pricing move, a positioning shift, a new segment). The failure mode is specific: an AI assistant fabricates a competitor fact (a Series B that never closed, a “10,000 customers” claim with no source), it lands in slide 4, and a VP repeats it in a board meeting. The whole job is a structured teardown with a deliberate verification step before anyone sees it.
Where AI helps — and where it lies
Treat the two halves of this task differently.
- Structuring (AI is strong). Given the homepage, pricing page, a few blog posts, and recent G2 reviews, a model will normalize messy text into a clean, consistent matrix faster than any human. This is its home turf.
- Recall (AI fabricates). Ask “how much funding has Acme raised?” with no source and the model returns a confident, invented answer. Studies of AI research tools consistently find that unsourced factual claims are where errors cluster.
The rule: every factual claim must trace to source text you provided. Any claim without a source line is fiction until you verify it. That single discipline is the difference between a useful teardown and an embarrassing one.
Which AI model for the research itself
If you do not want to paste pages by hand, the “deep research” modes will browse and cite sources for you. As of June 2026, here is how the three big assistants compare for this job:
| Tool | Plan to get deep research | Approx. limit | Best for competitor work |
|---|---|---|---|
| ChatGPT Deep Research | Plus $20/mo | ~25 runs/mo (Pro $200 ≈ 250) | Longest, most structured written reports; runs up to ~30 min |
| Gemini Deep Research | Google AI Pro $19.99/mo | High daily allowance; pulls from Gmail/Drive/Docs | Google-ecosystem workflows; report straight into a Doc |
| Claude (Opus 4.7 / research) | Pro $20/mo | Generous; indexes many sources per run | Wide source coverage and careful synthesis |
All three run on current top models — ChatGPT on GPT-5.5, Claude on Opus 4.7 / Sonnet 4.6, Gemini on Gemini 3.1 Pro — each with a 1M-token context standard, so you can paste a lot of competitor material in one shot. Pricing and limits here are time-sensitive; check the vendor page before you quote them. For a fuller breakdown see ChatGPT vs Claude vs Gemini.
Even with deep research, do not skip the paste step for anything load-bearing. Browsing modes still misattribute and occasionally cite the wrong company. The most reliable workflow is: let the model browse to find sources, then paste the exact text of the pages you will quote.
What to feed the AI
- Competitor names and URLs (homepage, pricing, blog, careers, changelog)
- Pasted page text or recent screenshots — not just links
- Your comparison dimensions, kept identical across all five competitors
- The specific decision this analysis informs (positioning move, pricing change, new market)
- Your own positioning, so the model can frame “what this means for us”
- Known facts and constraints (we are smaller, we are enterprise-only, we have no free tier)
Careers pages and changelogs are underrated. A competitor hiring six enterprise AEs or shipping an SSO/audit-log feature tells you more about their next move than the homepage copy does.
Copy-ready prompt
Build a structured competitor teardown.
Competitors and sources:
- [name] — [URL list] — pasted text: """[paste]"""
- ...
Dimensions to compare (identical across all competitors):
- Positioning (1 sentence)
- Pricing model and tiers
- GTM motion (PLG / sales-led / hybrid)
- Strongest signal of strength (with the evidence)
- Most visible vulnerability (with the evidence)
- Customer types (size, industry)
Our positioning: [line]
Decision this informs: [line]
Return:
1. A row per competitor on the listed dimensions
2. A 200-word "what this means for us" — what to do differently
3. A "watch list" — competitors to monitor closely, with a trigger for each
4. A fact-check log — every claim with the source line it came from.
If a claim has no source, flag it [UNVERIFIED]
5. The single counter-positioning move I should consider
Do not invent funding rounds, customer counts, ARR, or headcount.
Use only the pasted material. Where the material is silent, write "not stated".For a deep dive on one rival, follow up with: “Now write a one-page narrative on the strongest competitor — their core bet, the data they have that we don’t, and the single move that would hurt them most.”
Recommended output structure
A competitor matrix (rows = competitors, columns = dimensions), a one-paragraph “so what,” a watch list with triggers, and a fact-check log mapping every claim to its source. The fact-check log is not optional. Skip it and you ship hallucinations with a confident tone.
The fact-check pass
This is the step that makes the whole workflow safe to share:
- Scan the fact-check log for any
[UNVERIFIED]or “not stated” lines. Those are gaps, not facts — either remove the claim or go verify it. - Spot-check two or three of the most decision-relevant claims against the actual source page, not the AI’s paraphrase. Models compress, and compression introduces errors.
- For anything numeric (pricing tiers, employee count, raise amount), confirm it from a primary source. Pricing pages and the company’s own site beat a model’s memory every time.
- Date-stamp the teardown. Competitive facts rot; a deck with no date gets quoted six months later as if it were current.
When dedicated tools beat a chatbot
A general model handles the quarterly refresh fine. If you need continuous monitoring — alerts when a competitor changes pricing, ships a feature, or runs new ads — purpose-built competitive intelligence tools do that automatically. As of June 2026:
| Tool | What it does well | Rough cost |
|---|---|---|
| Crayon | Auto-tracks competitor sites, pricing pages, job posts, reviews; 4.6/5 on G2 | Mid five figures/year (~$15K–$30K+) |
| Klue | Battlecards for sales teams; 4.7/5 on G2; separate “curator” vs “consumer” seats | ~$16K+/year |
| Kompyte | Lighter mid-market monitoring with AI summaries and Slack alerts | ~$500–$1,500/mo |
| Similarweb | Traffic and digital-presence benchmarking; rare public pricing | From ~$125/mo (CI) |
The split is simple: a chatbot is the analyst that turns raw material into a teardown; a CI platform is the always-on sensor that tells you when the material changed. Many teams run both — Crayon or Kompyte for alerts, then an AI assistant to synthesize the quarterly story.
How to check the output is usable
- Every claim traces to a source line in the pasted material
- Dimensions are filled consistently — no “various / depends” in one row when others have a real answer
- The “what this means for us” suggests an action, not more analysis
- The counter-positioning move is real and specific, not “be better”
- The watch list has a trigger for each entry: what change would alter your decision
Common mistakes
- Comparing only on features — competitors win on GTM, pricing, and distribution as often as on features
- No “so what” — analysis without a decision is theatre
- Letting AI invent funding, customer count, or ARR (the single most common failure)
- Running it once a quarter and going dark — competitive shifts move faster; pair it with monitoring
- Freezing the dimensions forever — when the market changes, the comparison axes should too
FAQ
- Should I include indirect competitors? Yes, but name them once and move on. Their threat path is different, and the watch list is where they belong rather than the main matrix.
- How do I surface pricing that’s gated behind “contact sales”? AI cannot do this part. Get it from a sales call, a procurement contact, or a public RFP. Never let the model guess a number to fill the cell.
- What about open-source competitors? Use different dimensions — community size, governance model, contributor count, and dependency footprint matter more than tiered pricing.
- Can deep-research mode replace pasting sources? For finding sources, yes; for quoting them, no. Let it browse to locate pages, then paste the exact text of anything you’ll put in a slide.
- How often should I refresh? A full teardown quarterly, plus lightweight spot checks when a competitor raises money, ships a major feature, or changes pricing. Continuous-monitoring tools catch those triggers for you.
Related
- Market research organize AI — broader market research workflow
- Competitor comparison — focused 1-vs-1 comparison
- Business data analysis AI — adjacent business analysis
- Funnel analysis readout AI — your own funnel for context
- Feature prioritisation — turn analysis into roadmap
- AI industry research workflow — full industry research
- Excel / Spreadsheet Analysis Prompts