TL;DR: These 20 AI prompts for competitive analysis only work on evidence you paste in — a pricing page, exported reviews, a changelog entry, anonymized call notes — never on what a model recalls about a competitor. They cover positioning, feature gaps, pricing, messaging, reviews, changelogs, and win/loss data, ending in one prompt that synthesizes everything into a single decision.
How Do You Use AI Prompts for Competitive Analysis Without Inventing Competitor Facts?
You give the model evidence instead of a name. Ask "what does Competitor X charge?" and a language model answers anyway, filling the gap with a plausible-sounding number it cannot verify. Paste the competitor's actual pricing-page text and ask the same model to summarize, compare, or classify it, and the tool becomes reliable, because it's reasoning over text in front of it instead of guessing from training data that could be a year out of date.
That distinction, paste evidence versus recall a fact, is the whole difference between a competitive analysis you can act on and one that quietly embeds a wrong price or a feature the competitor dropped six months ago. Every prompt below names exactly what to paste for that reason.
Get this wrong in a sales battlecard or a board deck and the cost isn't a wasted afternoon. It's a rep quoting a competitor's price that changed last quarter, or a founder telling investors a rival lacks a feature it shipped in the spring. Both are the kind of specific, checkable claim that a prospect or an investor can verify in thirty seconds, and both are entirely avoidable if the number in the prompt came from a screenshot instead of a guess.
Why Does AI Confidently Get Competitor Facts Wrong?
Because a language model is optimized to answer, not to say "I don't know." OpenAI's own researchers put it plainly in a 2025 paper on the subject: "the training and evaluation procedures reward guessing over acknowledging uncertainty," since "language models are optimized to be good test-takers, and guessing when uncertain improves test performance" (source, accessed August 2026). A confident wrong answer scores better on most benchmarks than an honest "I'm not sure," so that's the behavior training reinforces.
Applied to a competitor's price or feature list, this means the model will produce a specific-sounding number even when it has no reliable basis for one, and that number reads exactly like every other sentence in the response — no hedge, no flag. Training data also has a cutoff, and pricing pages change without notice, so even a fact that was once true can be stale by the time you ask.
Picture two people asking the same question a week apart. Person A asks a chatbot, "does Competitor X have an API?" and gets a fluent, specific-sounding "yes, on their Pro plan," with no citation, because the model is pattern-matching against how similar companies at that stage usually structure their plans. Person B pastes the actual pricing page text and asks the same question. If the page doesn't mention an API, the model says so. If it does, the answer quotes the exact line. Same question, same model, and only one of those two answers is worth putting in a deck.
What Evidence Should You Gather, and Is This Ethical to Do?
Gather primary sources, not summaries of them. That means the competitor's own pricing-page text, copied directly rather than paraphrased from a third-party roundup, the feature list from their docs or product page, review text pulled from a public review site, a saved copy of their changelog, and, where you have it, anonymized sales-call notes mentioning them by name.
A third-party roundup or a "top 10 alternatives" listicle is not a primary source, even when it looks authoritative. Those pages get stale the moment a competitor updates a plan, and their author had the same access to public pages you do. Go to the competitor's own site for anything you'll act on, and treat a secondhand summary as a lead worth checking, not a fact worth pasting.
| Cluster | What you paste | What comes back |
|---|---|---|
| Positioning | Homepage hero copy, tagline, About page | Stated category, audience, and promise — stripped of hype |
| Feature gaps | Your feature list + their docs/pricing page | A sourced side-by-side, gaps marked "not stated," never "absent" |
| Pricing | Their pricing-page text | Tier structure, gated features, any unpublished or ambiguous price |
| Messaging | 3–5 real copy snippets | A voice description tied to specific lines, not a vibe |
| Reviews | Exported review text with dates | Recurring themes with counts, no invented sentiment score |
| Changelogs | Two saved snapshots | A diff of what actually changed between them |
| Win/loss | Call notes or transcript excerpts | The stated reason, in the prospect's own words, tagged as one data point |
Everything on that list is publicly visible to any prospect, which is also the ethical line worth naming directly.
How Do You Write a Prompt That Refuses to Guess?
Tell it to, explicitly. Every prompt below includes an instruction to flag anything the pasted text doesn't cover as "not stated," rather than let the model infer or fill the gap on its own. That one line turns a structured output from something that merely looks complete into something honest about its own limits, and it's the cheapest defense against hallucination available to you.
A weak prompt: "Compare our pricing to Competitor X's." A working prompt: "Compare our pricing to the text pasted below. If a plan, price, or feature isn't stated in the pasted text, write 'not stated' rather than estimate it." The second version is more boring to read and considerably harder to get wrong, which is the trade you want on anything you're actually going to act on.
Run both versions side by side once, on the same competitor, and the difference becomes obvious fast. The weak prompt fills every row of a comparison table, even the ones the pasted page never mentions. The working prompt leaves some rows marked "not stated," which looks like a less finished document until you remember that an unfinished-looking table beats a complete-looking one built partly on guesses. Treat every "not stated" you get back as a to-do item: go find that one fact, or accept that you genuinely don't have it yet.
What Are the Best AI Prompts for Competitive Analysis, by Task?
Each prompt below names what to paste. Fill in the brackets, paste the real evidence in place of the placeholder, and treat "not stated" in the output as a fact worth checking yourself, not a gap to fill in with what you already assume.
1–3. Positioning Teardown
Use these on a competitor's own words before you use anyone else's summary of them. Paste directly from their site, not a paraphrase you remember reading.
From the text pasted below only, extract:
1. The category they claim to be in (their own words, not your inference)
2. The stated target audience (quote the exact phrase if there is one)
3. The core promise or job they say the product does
4. Any claim in the text that isn't backed by a specific detail — flag it
as "unsupported claim," don't rewrite it into something concrete
Do not use anything you know about this company beyond what's pasted below.
If the text doesn't state something on this list, write "not stated."
[PASTE HOMEPAGE HERO COPY, TAGLINE, AND ABOUT PAGE TEXT]
Below are the homepage positioning statements for [OUR COMPANY] and up to
three competitors, pasted as-is.
For each, name only the category and audience explicitly stated or clearly
implied by the pasted text — not what you know about the company otherwise.
Then list where two or more overlap on category, audience, or promise, and
where each is genuinely distinct.
[PASTE OUR HOMEPAGE COPY]
[PASTE COMPETITOR A HOMEPAGE COPY]
[PASTE COMPETITOR B HOMEPAGE COPY]
Below is a features or benefits section from a competitor's site, pasted
as-is. For each stated benefit, ask "so what does that get the buyer" and
answer using only logic implied by the pasted text, not assumptions about
the product.
Sort the results into two groups: benefits with a concrete, specific payoff
stated in the text, and benefits that are adjectives with no measurable
payoff attached ("powerful," "innovative," "best-in-class").
[PASTE FEATURES OR BENEFITS SECTION]
4–6. Feature Gap Mapping
Feature gaps are product-manager work as much as marketing work — see 30 AI prompts for PRDs, user stories & roadmaps for the adjacent prompt set built for that role. These three turn two pasted feature lists into a gap map you can defend in a roadmap review.
Below are two feature lists: ours, and a competitor's, both pasted directly
from our own site and their public docs or pricing page.
Build a table with three columns: Feature, Us, Them.
Mark each cell Yes, No, Partial, or "not stated" — use "not stated" for
anything the pasted competitor text doesn't mention at all. Do not mark
anything "No" unless the pasted text explicitly says it's unavailable, or
the feature is simply absent from a documented, complete feature list.
[PASTE OUR FEATURE LIST]
[PASTE COMPETITOR FEATURE LIST FROM DOCS OR PRICING PAGE]
Below is the feature-gap table from the prompt above, plus a description of
our target user and the job they're trying to get done.
For each gap marked "No" or "Partial" on our side, state which specific job
from the persona below it blocks, if any. If a gap doesn't clearly block a
stated job, say so — don't invent a use case just to justify flagging it.
[PASTE FEATURE-GAP TABLE]
[PASTE TARGET USER / PERSONA DESCRIPTION]
Below is text from a competitor's public roadmap page, changelog, or recent
blog posts, pasted as-is.
Extract only features explicitly described as upcoming or planned. For
each, note whether the text frames it as: announced with a date, requested
by users (mentions of votes or "most requested"), or vague intent with no
commitment. Do not add anything you suspect they're building that isn't
stated in the pasted text.
[PASTE ROADMAP PAGE, CHANGELOG, OR BLOG POST TEXT]
7–9. Pricing-Page Analysis
Pricing pages are the single most-edited page on a competitor's site, so rerun these instead of trusting a screenshot from six months ago.
Below is the full text of a competitor's pricing page, pasted as-is.
Extract: every named tier, the stated price for each (or "not published" if
a tier says "contact sales" or similar), what's explicitly gated behind
each tier, and any billing detail mentioned (monthly/annual toggle, seat
minimums, overage fees). Flag anything ambiguous rather than resolving it
with a guess.
[PASTE FULL PRICING PAGE TEXT]
Using only the pricing-page text pasted below, classify the pricing model
(flat-rate, seat-based, usage-based, freemium, or a stated hybrid) and
identify any pricing psychology visible in the text itself: struck-through
"was" prices, a highlighted "most popular" tier, or a stated time-limited
discount. Quote the specific phrase that signals each one you flag.
[PASTE FULL PRICING PAGE TEXT]
Below are our own pricing details and a competitor's pricing-page text, both
pasted as-is. Team size to compare at: [NUMBER OF SEATS].
Calculate the total monthly cost for both at that team size, using only
stated prices and fees from the pasted text. If either page doesn't publish
a price at that team size (e.g., "contact sales" past a seat threshold),
say so explicitly instead of estimating a number.
[PASTE OUR PRICING DETAILS]
[PASTE COMPETITOR PRICING PAGE TEXT]
10–12. Messaging and Voice
Messaging audits are usually marketing-team work — best AI prompt tools for marketing teams covers the adjacent tooling question. These three work off real copy, never a paraphrase of it.
Below are 3-5 real snippets of a competitor's copy: a homepage line, one
email subject or body if you have it, one public social post.
Describe their voice using adjectives, but tie every adjective to the exact
line that supports it — for example, "direct: 'no free trial, no demo call,
just sign up'" rather than an unlinked adjective. If you can't tie an
adjective to a specific pasted line, drop it.
[PASTE 3-5 REAL COPY SNIPPETS WITH THEIR SOURCE — HOMEPAGE, EMAIL, SOCIAL POST]
Below is a competitor's homepage copy and our own ICP (ideal customer
profile) description.
Based only on the pasted homepage text, does the language, examples, and
implied reader match our ICP, a different ICP, or something broader than
either? Quote the specific phrases that support your answer.
[PASTE COMPETITOR HOMEPAGE COPY]
[PASTE OUR ICP DESCRIPTION]
Below is a single paragraph of landing-page copy — competitor's or our own.
Count how many concrete claims (a number, a named feature, a specific
outcome) appear versus how many unsupported adjectives ("powerful,"
"innovative," "world-class") with nothing measurable attached. Report both
counts and quote one example of each.
[PASTE ONE LANDING-PAGE PARAGRAPH]
13–15. Review Mining
Export or copy review text from a public review site — Chrome Web Store, G2, Capterra, the App Store — with dates attached. Never ask a model to recall what reviewers "generally say" about a product; it has no access to live review data.
Below are real reviews for a competitor's product, pasted with their dates
and star ratings.
Group them into recurring praise themes and recurring complaint themes.
For each theme, state how many of the pasted reviews mention it — don't
report a percentage or a sentiment score, only the count out of the
reviews actually pasted below.
[PASTE REVIEW TEXT WITH DATES AND RATINGS]
Below are the complaint themes extracted above, each with its review count.
For each complaint theme, state the specific product opportunity it points
to, tagged "if the complaint reflects current reality" since a review can
describe an older version of the product. Do not treat any single review
as representative of current sentiment.
[PASTE COMPLAINT THEMES WITH COUNTS]
Below are reviews with their dates, plus the approximate date of a known
product change: [DATE OF COMPETITOR'S LAST MAJOR UPDATE, IF KNOWN].
Split the pasted reviews into before and after that date, and report
whether the complaint themes differ between the two groups. If there
aren't enough reviews on one side of the date to say anything reliable,
state that directly instead of drawing a conclusion from too few reviews.
[PASTE REVIEW TEXT WITH DATES]
[DATE TO SPLIT ON]
16–17. Changelog Watching
Save a copy of a competitor's changelog page every time you check it. This comparison only works with two saved snapshots, not one live page plus your memory of what it said last time.
Below are two snapshots of the same competitor's changelog or release notes
page: one saved on [OLDER DATE], one saved on [NEWER DATE].
Summarize only what's present in the newer snapshot and absent from the
older one — new entries, not the full history repeated. Plain English, no
speculation about why they built it.
[PASTE OLDER CHANGELOG SNAPSHOT]
[PASTE NEWER CHANGELOG SNAPSHOT]
Below is a single changelog entry, pasted as-is.
Classify it as one of: catch-up feature (something competitors already
had), a genuinely new capability, a pricing or packaging change, or unclear
from the entry's wording alone. Base the classification only on how the
entry itself is worded, not on outside knowledge of their roadmap.
[PASTE ONE CHANGELOG ENTRY]
18–19. Win/Loss Framing
Anonymize before you paste. Strip the prospect's name and company from call notes or transcript excerpts, and keep the stated reason.
Below is a note or transcript excerpt where a prospect mentioned choosing or
rejecting [COMPETITOR NAME], with identifying details already removed.
Extract, in the prospect's own words as closely as possible: the stated
reason, and whether it was about price, a specific feature, timing, or
something else. Tag this as a single data point — do not generalize it into
a trend from one entry.
[PASTE ANONYMIZED CALL NOTES OR TRANSCRIPT EXCERPT]
Below are [NUMBER] separate anonymized win/loss entries, each already
structured with a stated reason and category.
If there are fewer than 5 entries, say the sample is too small to call
anything a pattern, and stop there. If there are 5 or more, report which
stated reasons repeat across how many separate entries, and note any
reason that appears only once.
[PASTE 5+ STRUCTURED WIN/LOSS ENTRIES]
20. Synthesis Into a Decision
This is the only prompt here that isn't paired with raw evidence directly. It runs on the saved outputs of the 19 prompts above, the same discipline our guide to prompt versioning for research applies to keeping dated versions of anything you'll revisit.
Below are my saved outputs from a subset of the prompts above: positioning
notes, the feature-gap table, the pricing breakdown, review themes,
changelog diffs, and win/loss patterns — whichever of these I've actually
run and saved.
Synthesize into a one-page brief with three sections: where we're actually
behind (cite which pasted output each claim comes from), where we're
actually ahead (same), and one recommended action for this quarter. Every
claim must reference which pasted section it's drawn from. If a section
below is missing, don't fill the gap with an assumption — note it as "not
assessed this round."
[PASTE SAVED OUTPUTS FROM PRIOR PROMPTS — LABEL EACH ONE]
How Do You Keep a Competitive Analysis Reliable Over Time?
Rerun on a cadence, not once. Pricing and positioning are worth checking quarterly, or the moment you notice a competitor's page has changed. Changelogs are worth a monthly pass, since that's roughly how often a serious competitor ships. Reviews and win/loss notes benefit from a standing weekly or monthly habit, so you're always comparing fresh evidence to fresh evidence instead of reasoning from a screenshot that's gone stale.
None of this requires new software. A saved library that treats each of the 20 prompts above as a reusable prompt template, ours or a plain notes doc with the same 20 prompts, keeps you from retyping the same "not stated" discipline into a blank chat window every quarter. If more than one person runs this — a co-founder, a marketer, a PM — a shared prompt manager, covered in best AI prompt managers for founders & small teams, keeps everyone pasting evidence into the same structure instead of drifting into their own shorthand.
Programs that track competitors this systematically report materially different outcomes: in Crayon's 2026 State of Competitive Intelligence report, teams running a formal program said nearly half, 49.6%, saw their win rate against competitors increase over the past year, versus 6.3% who saw it fall (source, accessed August 2026). Crayon sells competitive-intelligence software, so treat that figure as a directional, vendor-reported number rather than an independent audit. What seems to matter isn't the tool. It's running the analysis on a schedule, on real evidence, instead of once, from memory.
Before anything from these 20 prompts goes into a deck, an email, or a pricing decision, spot-check a handful of the specific claims against the source you originally pasted. Pick the two or three lines you'd be most embarrassed to have wrong in front of a customer, and confirm each one against the actual pricing page or review, not against your memory of running the prompt. That last check takes two minutes and it's the same discipline this whole guide opened with: the model can only be as accurate as the evidence it was given, and the only way to know it stayed accurate is to look.
Stop rewriting prompts. Start shipping.
Works with ChatGPT, Claude, Gemini, Grok, Midjourney, Ideogram, Veo3 & Kling. 5.0★ on the Chrome Web Store.
Create An Account