What you’ll be able to do
- Extract Voice of Customer (VoC) verbatim quotes (pain, objections, desired outcomes) with
sources
- Convert VoC into messaging angles (pain → promise → hook → headline → bullets)
- Map competitor messaging across ads, landing pages, and social (with URLs for
verification)
- Reduce boiled chicken by grounding decisions and copy in sources you can review
Grounded market truth
VoC → angles
Competitor map
Ship faster
Why this works
Most AI frustration is a grounding problem. Without trusted context, you tend to get:
- Cold start: you re-explain context and redo work every session
- Boiled chicken: output is “reasonable” but generic and bland
- Drunk uncle: the AI drifts, contradicts itself, or makes stuff up
Deep Research advantage: it returns research you can verify (quotes + URLs), so your next
prompts stay grounded.
The two Deep Research runs to do
Run these back-to-back. Together they produce (1) customer language and (2) category
positioning intelligence.
| Run |
Output (what you should demand) |
How you use it |
| Voice of Customer (VoC) |
- Table 1: verbatim quotes tagged by theme + sentiment, with source context
- Table 2: angle builder: pain → promise → hook → headline → bullets
|
Write copy that sounds like your market. Build campaigns around real pain and real objections. |
| Competitor Messaging Mapping |
A table of competitor hooks/angles/promises from ads, landing pages,
and social with URLs and verbatim capture.
|
See patterns, identify gaps, and clarify differentiation. |
Non‑negotiable: require URLs and request verbatim capture
wherever possible.
Run Deep Research (setup checklist)
- Draft the prompt in Google Docs so you can version and reuse it
- Enable Markdown (Tools → Preferences → enable Markdown)
- Copy as Markdown so headings/tables survive the copy
- Open ChatGPT (ideally inside your Marketing Project for organization)
- Turn on Deep Research and paste the prompt
- Answer clarifying questions (market, time range, preferred/avoided sources, competitor
list)
- Let it run (scope-dependent; larger prompts take longer)
- Ship something: pick one pain cluster/angle and generate an email, ad set, landing page
skeleton, or social series
Prompt rule: table-first outputs + one example row = predictable results.
From research → shipped marketing
The value shows up when you convert one insight into something publishable.
- Select one pain cluster (VoC) or one competitor angle you want to counter-position
- Anchor the model to that row (quote + source + angle) so output stays grounded
- Generate one asset from that anchor:
- Email storyboard (3–5 emails: subject → promise → bullets → CTA)
- Landing page skeleton (headline → subhead → bullets → objections → proof)
- Ad angles (3 angles × 3 hooks each)
- Social series (problem → proof → process)
FAQ
How long does Deep Research take?
Often 20–40+ minutes depending on scope (number of competitors, sources, and how broad the
prompt is).
Why do I still get generic output sometimes?
Usually the prompt is too broad, the expected output isn’t defined, or URLs/verbatim
capture weren’t required. Narrow scope and demand table-first outputs.
What sources should I include (or avoid)?
Include sources where real customers speak plainly (forums, communities, reviews) and where
competitors publish claims (ads/LPs/social). Exclude any source you don’t want inside the prompt.
What’s the most important ingredient in the prompt?
“What good looks like.” Define the tables/columns and show an example row.
Predictability goes up immediately.
Key terms (quick definitions)
- Deep Research: a slower, agent-style mode that searches broadly and returns a research
output based on your prompt
- Grounding: anchoring output to trusted sources (instructions + files + vetted inputs) to
reduce drift and hallucinations
- Voice of Customer (VoC): verbatim language from real people describing pain, objections,
and outcomes
- Messaging angle: a structured claim that connects pain → promise → hook/headline →
supporting bullets
- Competitor messaging map: a structured capture of competitor hooks/angles/promises with
URLs for verification
- Markdown: a formatting language that helps AI interpret structure (headings, tables) when
copying prompts
- Table-first output: designing prompts that return structured tables you can immediately
reuse (instead of long prose)
