How to Prompt AI for Competitive Research and Market Analysis

Ask ChatGPT who your competitors are and you get a confident list — half outdated, half hallucinated, rarely sourced. Competitive research prompting in 2026 requires source grounding, explicit [UNVERIFIED] flags, structured outputs (SWOT, feature matrices), and model routing that matches task type. Gemini 3.1 Pro handles large document corpora; Claude Opus 4.8 synthesizes nuanced positioning; GPT-5.5 produces structured JSON for slides. PromptMake /text offers 3 generations/day for guests and 5/day for registered users. This guide covers competitive research prompts, verification discipline, SWOT and feature matrix templates, source grounding architecture, Gemini/Claude routing, and FAQ.

Why open-ended competitor prompts fail

Weak: Analyze our competitors in the project management space. Output: names products that merged, features that never existed, market share percentages with no citation. Stakeholders paste into slides. Finance asks for sources. You have none. Strong: SOURCES block + ALLOWED INFERENCE rules + [UNVERIFIED] on every unsourced claim + JSON schema for matrix output.

Source grounding architecture

Paste or attach sources before the task. Order matters.

SOURCES (numbered):

[1] Competitor A pricing page — captured [date] — [paste or URL]

[2] Competitor B product docs — [date]

[3] Your internal battlecard — [date]

[4] G2 review summary you wrote — [date]

RULES:

• Every factual claim must cite [n] or [UNVERIFIED]
• Prefer primary sources over news aggregators
• If sources conflict, list both with [CONFLICT]
• Do not browse or invent sources not in SOURCES

No SOURCES → output only: INSUFFICIENT SOURCES — cannot complete factual analysis.

[UNVERIFIED] flag discipline

Prompt: Any statement not directly supported by SOURCES[n] must end with [UNVERIFIED].

Includes: pricing, feature availability, funding, headcount, market share, roadmap claims.

Analyst workflow: research brief is draft until every [UNVERIFIED] is resolved or deleted. Executives see clean brief; appendix keeps flagged items — not silent upgrades to fact.

ALLOWED INFERENCE scope

Define what reasoning is permitted:

ALLOWED INFERENCE:

• Comparative positioning language if both products described in sources
• SWOT themes labeled OPINION if strategic recommendation
• Feature matrix: Yes/No/Unknown only from source text — Unknown if not stated

NOT ALLOWED:

• Market size numbers without [n]
• Competitor intent ("they will likely...") without labeling SCENARIO:

Tight inference scope reduces confident fiction.

Competitive landscape overview template

TASK: Landscape overview for [your product] in [market]

SOURCES: [numbered list]

OUTPUT JSON:

{ "competitors": [{ "name": "", "segment": "", "positioning": "", "citations": [1] }], "unverified": [], "gaps_in_sources": [] }

Max 8 competitors. positioning must cite source or [UNVERIFIED].

SWOT prompt template

SUBJECT: [Competitor X or Your company]

SOURCES: [list]

OUTPUT:

Strengths: bullet + citation each

Weaknesses: bullet + citation each

Opportunities: label OPINION if not in sources

Threats: label OPINION if forward-looking

RULE: No bullet without [n] or [UNVERIFIED] or OPINION tag.

Claude Opus 4.8 for balanced SWOT nuance. Use for board-facing drafts.

Feature matrix template

TASK: Feature matrix — rows = features, columns = products

PRODUCTS: [Your product, A, B, C]

SOURCES: [docs, pricing pages]

CELL VALUES: Yes | No | Partial | Unknown

RULE: Yes/No/Partial require [n]. Unknown if not in sources — do not assume absence means No.

OUTPUT: markdown table + footnotes mapping cells to citations

GPT-5.5 excels at wide matrices with JSON backing.

Pricing comparison template

TASK: Pricing comparison table

SOURCES: official pricing pages only, dated

OUTPUT:

Tier | Price | Billing | Limits | Citation

If enterprise pricing undisclosed: Enterprise | Unknown | — | — | [UNVERIFIED]

Never estimate competitor pricing.

Positioning map prompt

TASK: 2x2 positioning description (not image) — axes [e.g. price vs depth]

SOURCES: [list]

OUTPUT: Quadrant placements with rationale + citations. Placements without source → [UNVERIFIED] placement.

Human draws slide visual — model supplies defensible placements only.

Win/loss analysis template

DATA: [paste anonymized win/loss notes or CRM export]

SOURCES: internal notes dated

TASK: Thematic analysis — why wins, why losses

OUTPUT JSON:

{ "win_themes": [{"theme":"", "count":0, "example_quote":"", "verify":true}], "loss_themes": [] }

Quotes must be verbatim from DATA or [PARAPHRASE].

Battlecard generation template

AUDIENCE: sales reps on calls

YOUR PRODUCT: [facts from internal source]

COMPETITOR: [name]

SOURCES: [battlecard inputs]

SECTIONS:

• When we win (cited)
• When we lose (cited)
• Landmines to avoid (OPINION ok)
• Questions to ask prospect
• Never say (compliance)

Max 1 page equivalent. Every competitor claim cited.

Market sizing — high risk zone

TASK: TAM/SAM/SOM

RULE: Only figures with [n] from reputable source in SOURCES. Otherwise output: MARKET SIZE: [UNVERIFIED — requires analyst research].

Do not prompt models for TAM without attaching analyst reports.

News and launch monitoring prompt

SOURCES: [press releases pasted with dates]

TASK: Timeline of competitor launches last 90 days

OUTPUT: Date | Event | Source [n] | Impact assessment [OPINION]

Separate fact timeline from impact opinions.

Gemini vs Claude routing

Gemini 3.1 Pro: ingest 10+ PDFs, earnings calls, long docs — feature extraction with citations

Gemini 3.5 Flash: quick first-pass source tagging — always re-run synthesis on 3.1 Pro or Opus

Claude Opus 4.8: strategic narrative, SWOT balance, sales battlecards, executive summary

GPT-5.5: JSON matrices, CRM exports, slide-ready bullet structure

Route by input size and output audience — not habit.

Multi-document Gemini workflow

Attach corpus to Gemini 3.1 Pro:

CORPUS: [competitor annual reports, docs]

TASK: Extract feature list per product with page/section citation

Then export to Claude Opus: Synthesize positioning narrative from EXTRACT below — preserve citations.

Two-step reduces single-model hallucination on long docs.

Multi-model verification pass

Pass 1: Gemini 3.1 Pro extracts facts with citations from corpus. Pass 2: GPT-5.5 builds matrix JSON from extraction only — no new facts. Pass 3: Claude Opus 4.8 writes narrative and flags contradictions. Three-model pipeline cuts single-model hallucination on competitive decks. Budget analyst time for the pipeline — not one-shot prompting.

Claude synthesis workflow

Paste Gemini extraction + SOURCES into Claude Opus:

EXTRACT: [structured facts with [n]]

TASK: Executive summary 300 words, no new facts, flag any contradiction

Opus catches narrative inconsistency Gemini may miss.

Executive summary template

AUDIENCE: CEO, 5-minute read

SOURCES: [list]

OUTPUT:

• Situation (cited)
• Top 3 competitor moves (cited)
• Top 3 implications for us (OPINION labeled)
• Recommended actions (OPINION)
• Open questions requiring research

Max 400 words. No [UNVERIFIED] in main bullets — move to appendix.

Sales objection handling

OBJECTION: [prospect says competitor cheaper]

SOURCES: [pricing, TCO doc]

OUTPUT: response framework with cited facts + questions back — no trash talk

Label OPINION on value framing.

Product roadmap competitive response

COMPETITOR ANNOUNCEMENT: [paste from source]

YOUR ROADMAP (internal): [paste]

TASK: gap analysis — what we have, what we lack, messaging DO/DON'T

OPINION on messaging. Facts on features must cite internal or external source.

Porter five forces prompt (optional framework)

TASK: Five forces analysis for [market]

SOURCES: [numbered — industry reports only]

OUTPUT: Force | Assessment | Evidence [n] or [UNVERIFIED] | Implication [OPINION]

Forces: rivalry, new entrants, substitutes, buyer power, supplier power

RULE: Industry-level claims require [n]. Company-specific actions labeled OPINION.

Earnings call and investor doc extraction

CORPUS: [earnings transcript excerpt]

TASK: Extract competitor-mentioned metrics, guidance quotes, product roadmap hints

OUTPUT: Quote | Speaker | Page/section | Strategic implication [OPINION]

Gemini 3.1 Pro on long transcripts. No paraphrase as direct quote.

Patent and IP scan disclaimer

TASK: Summarize publicly listed patents if in SOURCES

RULE: Not legal advice. No claim of infringement. [UNVERIFIED] if patent status unclear.

OUTPUT: Patent ID | Summary | Relevance to our product [OPINION] | citation

Route IP questions to counsel — not sales battlecards without review.

Ethical and legal boundaries

Do not prompt for: employee PII, stolen documents, wiretaps, misrepresentation. Public sources, licensed reports, internal win/loss only. No defamatory statements. Competitor criticism must be factual and cited.

Scenario planning template

SCENARIO: [Competitor drops price 20% / launches AI feature] | SOURCES: current pricing, roadmap | OUTPUT: Assumptions [OPINION], impact on us [OPINION], messaging DO/DON'T [OPINION], facts with [n] only | Label entire section SCENARIO — not prediction.

Win-loss interview synthesis

DATA: [paste 5–10 anonymized loss call notes]

TASK: extract recurring objections, competitor mentions, decision criteria

OUTPUT JSON: objections[], competitor_citations[], recommended_enablement [OPINION]

Claude Opus 4.8 for qualitative synthesis. Every competitor attribute in notes gets [INTERNAL] tag.

Regional and regulatory market overlays

MARKET: [EU / US / APAC] | SOURCES: local competitor sites in language | TASK: pricing and feature matrix for regional variants | RULE: [UNVERIFIED] if source not locale-specific | Gemini 3.1 Pro for multilingual paste with per-paragraph citation.

Gartner and analyst report summarization

SOURCE: [licensed analyst excerpt — paste only what license allows]

TASK: extract vendor rankings and quadrant positions with page citation

RULE: no redistribution of paid report text in full; summary bullets only; [UNVERIFIED] if excerpt incomplete

OUTPUT: finding | citation | relevance to our positioning [OPINION]

Refresh cadence

Competitive briefs stale fast. Tag every output AS OF: [date] and SOURCE CAPTURE DATES. Flag sources older than 90 days as [STALE SOURCE] in gaps_in_sources.

Before vs after: competitor list

Weak: Who are our competitors?

Output: 10 names, 3 wrong category, 2 acquired.

Strong: SOURCES + landscape JSON + unverified array empty or explicit.

Before vs after: feature matrix

Weak: Compare us to Competitor A features.

Output: checkmarks for features A does not ship.

Strong: Yes/No/Unknown with citations; Unknown until confirmed.

Contradiction handling

If [2] says API included and [3] says add-on:

OUTPUT: [CONFLICT] API pricing — [2] states X, [3] states Y — analyst resolve

Models should surface conflict — not pick a side silently.

Internal vs external sources

INTERNAL: win/loss, roadmap — label [INTERNAL]

EXTERNAL: pricing page — label [n]

Prompt: Do not present INTERNAL roadmap as public competitor fact.

Competitive monitoring cadence and tooling

Weekly: pricing page capture into SOURCES numbering scheme. Monthly: feature matrix refresh. Quarterly: SWOT and positioning narrative for leadership. Tag outputs AS OF date. Integrate CRM loss reasons into win/loss template — internal [INTERNAL] citations only. PromptMake /text for matrix and SWOT on pasted extracts; Gemini 3.1 Pro for full PDF corpus outside daily generation limits.

Analyst QA checklist

Checklist before publish: competitor names approved; zero unsourced numbers in exec summary; matrix Unknowns investigated; [STALE SOURCE] refreshed; OPINION separated from fact; [UNVERIFIED] appendix complete.

Common failures

No sources → fabricated market share. Assuming No in matrix when feature not mentioned → false negatives. Single-model long PDF → missed citations. Removing [UNVERIFIED] for slide aesthetics → misinformation. Battlecard trash talk → legal exposure. TAM from model memory → board-level error.

JSON schema for slide pipelines

{ "matrix": [], "swot": {}, "citations": {}, "unverified": [], "opinions": [], "as_of": "" }

GPT-5.5 structured output feeds Google Slides scripts or Notion.

FAQ

Can AI replace competitive intelligence research?

No. AI accelerates synthesis and structuring. Humans source, verify, and decide. [UNVERIFIED] exists because models guess.

When is [UNVERIFIED] mandatory?

Any claim not directly supported by a numbered SOURCE in the prompt. Default to flag — not exception.

Best model for 50-page competitor PDF?

Gemini 3.1 Pro for extraction with citations. Claude Opus 4.8 for executive narrative from extraction.

Feature matrix — absence of mention?

Mark Unknown, not No. Confirm absence from official docs or sales engineering.

How to handle stale pricing pages?

Tag [STALE SOURCE]. Re-capture page before battlecard publish.

SWOT Opportunities — need citations?

Label OPINION unless sourced market data. Opportunities are often strategic inference.

Can I use ChatGPT browsing for competitors?

Browsing adds sources — still require capture date and cite URL in SOURCES. Do not mix unattached browsing with grounded prompt in same pass without listing URLs.

GPT-5.5 vs Opus for battlecards?

Opus for nuance and sales tone. GPT-5.5 for strict JSON + citation fields.

How many competitors per prompt?

3–5 for matrix quality. Eight max in landscape JSON. More → split runs.

PromptMake limits for research sprints?

Guests 3 generations/day, registered 5. Use Gemini outside PromptMake for huge corpus; PromptMake for matrix/SWOT templates on pasted extracts.

Related articles

ai-prompts-data-analysis-structured-insights, prompting-ai-legal-summaries, claude-vs-chatgpt-vs-gemini-text-prompts — verification flags, source discipline, model routing

Bottom line

Competitive research prompting demands SOURCES, [UNVERIFIED] discipline, and structured outputs — SWOT, matrices, battlecards. Gemini 3.1 Pro for corpus, Claude Opus 4.8 for narrative, GPT-5.5 for JSON. Unknown beats false No. Analyst reviews every brief before slides. PromptMake guests get 3 /text runs daily; registered users get 5 — use them for matrix and SWOT templates on pasted extracts, not raw guessing.