TL;DR
- The job: synthesize messy research (interviews, teardowns, reviews) into a 15-minute landscape doc with named segments, one real gap, and a kill-switch assumption your VP can greenlight or veto.
- Tool fit (June 2026): paste-and-synthesize fast in Claude Pro ($20/mo, Sonnet 4.6, 1M-token context, Projects knowledge base); use NotebookLM (free, 50 sources) when you need every claim citation-backed to a source line; use ChatGPT Plus ($20/mo, GPT-5.5) if it is already your daily driver, but watch the ~320-page in-app context ceiling.
- The trap: the confirming landscape — AI faithfully builds a map that validates the strategy you walked in believing. The fix is in the prompt below: force behavior-based segments and demand the disconfirming evidence in your own notes.
- Non-negotiable line: “What assumption, if wrong, kills this recommendation?” Cite the note that already hints it might be wrong.
The task
You have 4 competitor teardowns, 6 customer interview transcripts, a 20-page Notion of scraped review excerpts, and a half-finished Miro board. Strategy committee meets Thursday. They want a clear landscape doc: what the segments are, who’s playing in each, where the real gap is, and the one thing that, if you’re wrong about it, kills the whole entry recommendation. Not a 40-page deck. A doc your VP can read in 15 minutes and either greenlight or veto.
Where AI helps — and where it does not
AI is excellent at synthesizing a pile of unstructured notes into a clean landscape (segments, competitor maps, gap candidates) and at flagging contradictions between two interview quotes that you didn’t notice. It is also good at converting “they said something about pricing” into “3 of 6 interviewees flagged the per-seat pricing as a blocker over 10 seats.” What AI cannot do: validate your positioning. If your input positioning is wrong, AI will faithfully build a landscape that confirms it, segment by segment. Run the kill-switch question explicitly: “what assumption, if wrong, kills this recommendation?”
The named failure mode: the confirming landscape. AI gives you a beautifully structured landscape that happens to validate the strategy you walked in believing. The cure is to ask for segments defined by customer behavior, not by your product, and to require AI to name the disconfirming evidence in your own notes.
Which tool, and why (June 2026)
The job is really two things: (1) hold a big pile of unstructured notes in context at once, and (2) never invent a quote. Those two needs pull toward different tools.
| Tool | Plan / price | Context for raw notes | Why pick it for this | Watch-out |
|---|---|---|---|---|
| Claude Pro | $20/mo ($17 annual) | Sonnet 4.6, 1M tokens standard; Projects knowledge base ~200K with RAG | Best default. Paste 6 transcripts + teardowns directly into a Project, then run the prompt across multiple sessions without re-pasting. Strong at holding contradictions in view. | Heavy synthesis can hit Pro message caps; upgrade to Max ($100) for sustained all-day runs. |
| NotebookLM | Free (50 sources); Plus 300 sources, 500K words/source | Closed-RAG, strictly grounded in your uploads | When “do not invent quotes” is the priority. Every sentence carries an inline citation back to the source line, so the kill-switch evidence is auditable. | Won’t reason as freely across your hypotheses; it answers, it doesn’t argue. Use it to verify, Claude to synthesize. |
| ChatGPT Plus | $20/mo | GPT-5.5; in-app ~320 pages (full 1M only on $200 Pro) | Fine if it’s already your daily driver. GPT-5.5 Thinking mode is sharp on the kill-switch logic. | The ~320-page in-app ceiling bites first on long-transcript piles. Upload 20 files max per message (raised Feb 2026); a 50-page PDF alone can eat 30K+ tokens. |
| Gemini (AI Pro) | $19.99/mo | Gemini 3.1 Pro, 1M tokens + Workspace | Good if your notes already live in Google Docs/Drive; it reads them in place. | Less disciplined about refusing to fill gaps; lean on the “say missing” rule below. |
Practical pattern most teams land on: synthesize in Claude, verify the kill-switch evidence in NotebookLM. Two models disagreeing on the kill-switch is itself signal (see FAQ). Specs cited here are from the vendors’ own pages — Anthropic on Claude context and Projects and Google’s NotebookLM help center — and shift often, so re-check before you commit budget.
What to feed the AI
- Raw notes (interview transcripts, review excerpts, competitor teardowns, pricing pages, app store reviews)
- The single strategic question (“should we enter segment X?”, “where can we grow market share?”, “which sub-segment of Y has the weakest incumbent?”)
- Your company’s current positioning (one sentence, the way a sales rep would say it)
- The 2-3 hypotheses you walked into the research holding
- Any segments you already think exist (so AI can confirm or break them)
- The decision being made on Thursday (kill / build / expand / hold / partner)
- Who reads the doc (VP, board, PMM team, sales). Each needs a slightly different surface
- The piece of disconfirming evidence in your notes that you noticed and brushed past
Copy-ready prompt
You are synthesizing market research into a landscape doc.
Strategic question: {one sentence — should we / where should we / which one}
Our positioning (one sentence as a sales rep would say it): {paste}
Hypotheses I walked in with (so you can confirm or break them): {2-3 hypotheses}
Decision being made: {kill / build / expand / hold / partner}
Audience: {VP / board / PMM / sales}
Raw notes (interviews, reviews, competitor pages):
{paste all the raw notes}
Synthesize as:
1) Market segments — define each by *customer behavior or job-to-be-done*, not by demographics or by our product. 3-5 segments max.
2) For each segment: top 2-3 real, named competitors (with one-line positioning each); the dominant pricing model; the most common reason customers in this segment churn or fail to adopt.
3) The gap we could exploit — specific, not "be better than X." Name the underserved behavior or unsolved job.
4) Biggest risk if we enter — phrased as a thing we should watch for after entry, not as a hand-wave.
5) Recommendation — kill / build / expand / hold / partner — with one paragraph of reasoning.
6) Kill-switch assumption — the single assumption that, if wrong, kills this recommendation. Cite the line in my notes that already hints it might be wrong.
7) Disconfirming evidence I gave you — list 2-3 quotes from my notes that contradict my hypotheses. If none, say so explicitly (this is a red flag).
Rules:
- Do not invent quotes or competitors. If a fact is missing, say "missing."
- Segments defined by behavior, not by product features.
- Never write "be better than" or "differentiate by quality."
- The kill-switch assumption is mandatory.Shorter variant — landscape snapshot only
Give me a 1-page landscape snapshot from these notes.
Question: {one line}. Positioning: {one line}.
Output: 3 segments × (top 2 competitors + dominant pricing + biggest gap).
At the bottom: the one kill-switch assumption. 250 words total.
Notes: {paste}Sample output
A useful segment definition: “Solo founders building B2B SaaS pre-seed (n=2 founders, no engineering hire, runway <12 months). Dominant behavior: they buy tools at the moment they hit a specific pain — usually after losing a customer to it — and abandon tools that require >30min to set up. They are not ‘small business owners’; the buying behavior is fundamentally different.”
A useful gap statement: “Underserved behavior: pre-seed B2B founders need a billing tool but cannot justify a per-seat license when there is one founder, and the existing ‘free up to N customers’ tiers all gate the integrations they need on day one. Stripe handles the payment layer but not the lifecycle. The gap is a flat-fee, low-floor billing layer for solo founders, not a feature competition.”
A useful kill-switch: “Kill-switch assumption: that solo founders will pay $29/mo flat once they cross 10 customers. If they stay free past that threshold, segment LTV won’t support CAC. Evidence in our notes pointing to risk: 4 of 6 interviewees said they would ‘try to stay on the free tier as long as possible’ (Acme interview line 47, Beta interview line 31, Gamma interview line 12, Delta interview line 19).”
How to refine
- Force behavior-based segments: “Redefine the segments using customer behavior or jobs-to-be-done. If two segments share the same behavior, merge them. If one ‘segment’ is actually a demographic, replace it or delete it.”
- Disconfirm yourself: “List the 3 strongest quotes from my notes that contradict my hypotheses. Do not soften them. If you cannot find 3, tell me which interviews I need to redo.”
- Make the gap specific: “Replace ‘be better than X’ with the underserved behavior or the unsolved job. Cite the quote in my notes that names it.”
- Name the watch-list: “For the ‘biggest risk if we enter,’ rewrite it as the 2 things I should watch in the first 6 months — leading indicators, not lagging.”
- Pressure-test the kill-switch: “Name the second kill-switch assumption — the one I would notice fail second, after the obvious one. Often the second kill-switch is what actually breaks the strategy.”
Common mistakes
- Asking AI to synthesize without supplying your positioning. It picks one for you, usually a generic one
- Skipping the kill-switch question. Landscape feels rigorous, strategy fails 6 months later for a reason nobody named
- Segments defined by demographics (size, geography, industry) when behavior is the actual variable
- Generic competitor list with no positioning per competitor. “Acme, Beta, Gamma” means nothing without their one-line stance
- “Gap” expressed as “be better than X”. That’s not a gap, that’s a brand statement
- One-shot landscape that doesn’t get re-validated against reality 90 days after entry. The landscape ages fast in active markets
- Letting AI invent quotes or attribute statements to a competitor that did not actually say them. If a fact is missing, mark it missing
- No disconfirming evidence listed. If the notes have nothing that contradicts your hypothesis, you under-interviewed
FAQ
- What if my notes are messy?: Better than tidy notes that have already been filtered. Paste verbatim. AI will surface the contradictions you flattened during note-taking.
- Will all my notes fit in one prompt?: Usually yes in Claude Sonnet 4.6 or Gemini 3.1 Pro (1M-token context as of June 2026 — roughly 700K words, far more than any teardown pile). ChatGPT Plus is tighter: the in-app context is about 320 pages, so for big transcript stacks either upload as files (20 per message) or split into two passes. NotebookLM sidesteps the limit entirely by retrieving from up to 50 sources.
- Should I run this with multiple LLMs?: Yes for high-stakes recommendations. Run the same prompt through Claude Opus 4.7 and GPT-5.5 Thinking; they surface different gaps and often disagree on the kill-switch, which is signal. If both name the same kill-switch, you’ve probably found the real one. NotebookLM is the tiebreaker because it cites the source line.
- How do I stop it inventing quotes?: Keep the “if a fact is missing, say missing” rule in the prompt, and for anything that goes in front of the board, verify the quote in NotebookLM, which grounds every sentence in a cited source passage. Gemini and ChatGPT are the most likely to smooth over a gap with a plausible-sounding fabrication.
- How often to refresh?: Quarterly for active markets (new entrants, shifting pricing). Semi-annually for stable ones. Always re-validate after a quarter post-entry. The landscape that justified entry is not the same one you’re operating in.
- What if the recommendation is “kill”?: Lead with kill, write it crisp. A confident kill saves a year. Audiences are bad at reading “leaning kill but with caveats”; they round it up to “build.”
- Should I share raw transcripts or summaries?: Raw, when you can. Summaries already encode your bias. Strip names if needed; keep the language verbatim. It’s where the disconfirming signal lives.
Related
- Competitor comparison — 1v1 deep dive
- Competitor analysis AI — structured teardown
- Market research summary — lighter summary
- Business data analysis AI — adjacent business analysis
- Feature prioritization — translate landscape into roadmap
- Roadmap planning AI — once positioning is clear
- Company Research Prompts: 12 Templates for Interview-Ready Background
Tags: #AI writing #Data analysis #Finance #Research #Competitor