Product Marketing Alliance · AI Show and Tell: Marketing Edition · Builder's Club · September 23, 2026
AI for Marketers.
These are the prompts from the workshop, exactly as they were run. Copy them, swap in your own competitor, and you've got the same thing running on your category in about ten minutes.
Instructor Heather Gawel · Momentum Product Co.
Host Product Marketing Alliance
Session Competitive Agent Workflow
Take the prompts with you
Get the prompts.
The four-step competitor gap chain, the one-prompt version, and a worked example of what to paste into step 4. Plus the occasional note from Momentum on what we're building next.
Four prompts · run in order
The competitor gap chain
Four prompts, run in order. The first two collect what a competitor claims and what their buyers actually say, kept strictly apart. The third finds the distance between them. The fourth turns those buyers into personas and points them at your own copy.
The separation is the part that matters. A model that's already read the marketing goes looking for agreement.
Step 1 — what they claim
Research Pendo (pendo.io), a product adoption and onboarding platform.
Collect ONLY what Pendo says about itself: homepage headline and subhead,
the claims on its product and pricing pages, its published proof points
and customer numbers.
Rules:
- Quote exactly. No paraphrasing.
- No third-party commentary, no reviews, no analyst copy.
- Note which page each claim came from.
Output 8-12 claims verbatim.
Step 2 — what their buyers say
Now collect what Pendo's actual customers say, from review sites (G2,
Capterra, TrustRadius), Reddit, and community threads.
Rules:
- Real quotes from real users only. If you can't find a real quote for
something, say so. Do not invent, compose, or smooth out a quote.
- Keep this pile completely separate from step 1. Do not compare yet.
Do not look for agreement with the marketing.
- Prioritise reviews that mention: implementation and time to value,
pricing and contract experience, who owns and maintains the tool,
and reporting or proving results.
For each quote give: the quote, the reviewer's role and company size if
stated, and the source.
Aim for 15-20 quotes.
Step 3 — the gap
Now compare the two piles. Produce three things.
a) A table:
Claim | What buyers actually say | Verdict (holds / partial /
contradicted) | The quote that decides it
b) A silence list: claims Pendo makes that NO buyer quote mentions at
all. Treat this as just as important as the contradictions — a claim
nobody repeats is a claim nobody heard.
c) The three sharpest single quotes, the ones worth reading aloud.
Step 4 — the war game
Build 3 buyer personas strictly from the buyer voice you collected in
step 2.
Each persona must be traceable: role, company size, and the 2-3 direct
quotes they are built from. Do not add motivations, budgets or
priorities that no quote supports.
Then show each persona the messaging below and have them react in their
own voice — sceptical, specific, and in the register of the reviews they
came from. Not a focus group. A buyer who has been burned before.
[PASTE YOUR OWN HERO COPY, YOUR THREE CORE CLAIMS, YOUR PRICING POSTURE,
AND — THIS IS THE PART THAT MATTERS — WHAT IS TRUE INTERNALLY THAT YOU
DO NOT PUBLISH]
For each persona, give:
- First reaction, in their voice
- The specific objection
- What they would need to see before believing the claim
- A verdict on the current messaging: yes / no / not yet
Then close with:
- Which claim survives all three personas
- Which claim dies
- The one thing all three are asking for, and whether our messaging
actually claims it, gestures at it, or misses it
- Whether we can back that claim up today, judged against what is true
internally and not the public proof points. If we cannot, say so
plainly and say what would have to be true first
Be hard on us. A persona that likes everything is a persona you built
wrong.
One paste
The whole thing in one prompt
Same workflow, one paste, if you'd rather not babysit four steps. Swap [COMPETITOR] for whoever you're up against.
The whole thing in one prompt
You are doing competitive positioning research. Work through four steps
in order. Show your full output for each step before moving to the next.
STEP 1 — WHAT THE COMPETITOR CLAIMS
Research [COMPETITOR], and collect ONLY what they say about themselves:
homepage headline and subhead, the claims on their product and pricing
pages, their published proof points.
Quote exactly. No paraphrasing. No third-party commentary, no reviews.
Note which page each claim came from. Output 8-12 claims verbatim.
STEP 2 — WHAT THEIR BUYERS SAY
Now collect what their actual customers say, from review sites (G2,
Capterra, TrustRadius), Reddit, and community threads.
Real quotes from real users only. If you can't find a real quote for
something, say so. Do not invent, compose, or smooth out a quote.
Keep this pile completely separate from step 1. Do not compare yet. Do
not look for agreement with the marketing.
Prioritise reviews that mention implementation and time to value,
pricing and contract experience, who owns and maintains the tool, and
reporting or proving results.
For each quote: the quote, the reviewer's role and company size if
stated, and the source. Aim for 15-20.
STEP 3 — THE GAP
Compare the two piles. Produce:
a) A table: Claim | What buyers actually say | Verdict (holds /
partial / contradicted) | The quote that decides it
b) A silence list: claims they make that NO buyer quote mentions at
all. A claim nobody repeats is a claim nobody heard.
c) The three sharpest single quotes, the ones worth reading aloud.
STEP 4 — THE WAR GAME
Build 3 buyer personas strictly from the buyer voice in step 2. Each
must be traceable: role, company size, and the 2-3 direct quotes they
are built from. Do not add motivations no quote supports.
Then show each persona this messaging and have them react in their own
voice — sceptical, specific, in the register of the reviews they came
from. Not a focus group. A buyer who has been burned before.
[PASTE YOUR HERO COPY, THREE CORE CLAIMS, PRICING POSTURE, AND WHAT IS
TRUE INTERNALLY THAT YOU DO NOT PUBLISH]
For each persona: first reaction, the specific objection, what they'd
need to see, and a verdict on the messaging (yes / no / not yet).
Then close with:
- Which claim survives all three personas
- Which claim dies
- The one thing all three are asking for, and whether our messaging
claims it, gestures at it, or misses it
- Whether we can back it up today, judged against what is true
internally and not the public proof points. If we cannot, say so
plainly and say what would have to be true first
Be hard on us. A persona that likes everything is a persona you built
wrong.
Worked example
What to paste into step 4, worked through
Step 4 asks for your messaging, and this is where most people get a polite rewrite instead of something useful. The trick is to hand over what's true inside the company alongside what's on the website.
Here's a filled-in example so you can see the shape. The company is made up. The structure is the real thing.
What to paste into step 4
OUR MESSAGING (Launchpad — a fictional company, used as an example)
Hero: "The system of record for customer onboarding"
Subhead: "Build the onboarding your customers actually complete — and
prove to your board it worked. Guided flows, audit trail, time-to-value
you can show."
Core claims:
1. Prove it worked — every flow, every completion, every customer who
reached first value, in a report you can hand to leadership.
2. Shorter time to first value — the one metric your renewal depends
on, and the one we're built to move.
3. Built for whoever inherits it — flows outlive the person who built
them.
Pricing: three tiers. Starter self-serve, Growth mid-market, Scale
enterprise with SSO, SCIM and audit logging. Scale is "talk to sales."
Public proof points: ~340 customers, SOC 2 Type II. No published
time-to-value number. No case study showing a customer shortened theirs.
True internally, not published:
- Annual logo churn is 14%; the 2026 goal is under 10%
- "Customers cannot currently show their own leadership that Launchpad
worked" — our own words, and the most cited reason in closed-lost
- Analytics is mid-re-platform until Q1 2027
- Permissions and sharing are frozen until Q2 2027
- No native mobile authoring in 2026
- No professional services org, and none planned
Judge our claims against what is true internally, not the public proof
points.
The last section is the one that does the work. Without it you'll get a nicer version of what you already wrote. With it, you get told which claims you haven't earned yet — which is the part worth knowing.