Extract insights from millions of real AI conversations

People tell AI what they need. Observatory turns it into the patterns, prompts and personas behind every decision.

$ curl https://api.observedintelligence.com/v1/ask \

-H "Authorization: Bearer $OBSERVATORY_KEY" \

-d '{"question": "How do small business owners ask AI to choose a CRM?", "market": "US"}'

Response · 200

{

"answer": "Most never say “CRM”. They describe a mess.",

"sample": 38412,

"representative_prompts": [

"my sales guy keeps every lead in whatsapp and we lose half of them. what should a 6 person company use",

"is the free plan actually free or do they make you pay once you grow"

"drivers": [

{"name": "Cost as the team grows", "share": 0.41, "ci": [0.38, 0.44]},

{"name": "Setup in a day", "share": 0.33, "ci": [0.30, 0.36]},

{"name": "Works with WhatsApp and email", "share": 0.29, "ci": [0.27, 0.31]}

]

}

Observatory is a research agent over hundreds of millions of consented, anonymized AI conversations. Ask about any market. Get the prompts, the reasons and the personas behind it.

Early access Talk to us

For product marketers, founders, researchers and AI product teams.

Every decision today starts in AI chat Until now, nobody could see them

What people want appears in chat before it becomes a choice. It’s the clearest record of intent ever created.

What teams assumebest crm small business4 words

What people write

my sales guy keeps every lead in whatsapp and we lose half of them. what should a 6 person company use that isn’t expensive

24 words

Context6 person company
ProblemLeads lost in WhatsApp
ConstraintNot expensive

What teams assumebest sunscreen dark skin4 words

What people write

spf for dark skin that doesn’t leave a white cast and isn’t greasy under makeup, under $20

17 words

ContextDark skin, wears makeup
ProblemWhite cast, greasy finish
ConstraintUnder $20

What teams assumecar insurance renewal3 words

What people write

My renewal costs more this year and I haven’t claimed. How should I compare a cheaper policy with a higher excess?

21 words

ContextPrice up, no claims
ProblemHow to compare policies
ConstraintCheaper, with a higher excess

What teams assumevector database comparison3 words

What people write

pgvector or Qdrant if I need per-tenant filters and we’re already on Postgres?

13 words

ContextAlready on Postgres
Problempgvector or Qdrant
ConstraintPer-tenant filters

Turn AI conversations into decisions you can act on

Ask the questions you would usually commission a study for. Every answer comes from what people actually wrote.

Phrasing

How people describe the need in their own words, by market and language.

Sunscreen · US · ENRepresentative

“spf for dark skin that doesn’t leave a white cast and isn’t greasy under makeup”

“white cast”in 31% of these conversations

“is mineral sunscreen actually better or is that marketing”
“my 4 year old has eczema, which sunscreen won’t sting”
“something for the gym that doesn’t run into my eyes, under $20”
“do i still need spf if my foundation already has some”
“which sunscreen can i reapply over makeup at work”
“tinted spf that matches deep skin and doesn’t look ashy”

Decision drivers

The criteria, limits and deal-breakers people state before they choose.

Alternatives

What people weigh against each other, and why one wins.

CRM · small teamsReasons, by option

Option	Chosen when	Dropped when
Spreadsheet	It is just the founder	Leads get lost
Free plan	Under five people	Price jumps with seats
Paid CRM	The team is growing	Setup takes weeks

Friction

The confusion, objections and follow-up questions that slow a decision down.

Personas

Grounded in observed conversation patterns. Not simulated respondents.

Meal kits · 3 personasObserved

The weeknight optimizer

cheapest meal kit for 2 that takes under 30 min

Decides on: Prep time, price per serving
Worries: Too much packaging

The new cook

easy dinners for someone who can’t cook and lives alone

Decides on: Step-by-step clarity
Worries: Wasting food

The macro counter

high protein meal kit under 600 calories a serving

Decides on: Nutrition per meal
Worries: Hidden sugar

Where they differ

Time pressure

Price focus

Cooking skill

Diet focus

LowHigh

Built from patterns across 12,906 conversations. No individual is described.

Trends

How the questions shift over time, across languages and markets.

Prompt sets

Export prompt sets that read like your users, tagged by intent and stage. Test your AI product or write evals.

Prompt	Intent	Stage
how much down payment do i actually need for a first home	Budget	Research
should i buy now or wait for rates to drop	Timing	Consider
can i get a mortgage if i’ve been self employed for 18 months	Financing	Research
what does a home inspection actually check	Process	Decide

first-home-buyer.jsonl

2.4kprompts

EN, ES · intent and stage tags included observatory · search_prompts

Written the way people write In the languages they think in

Long, specific, often in their first language.

However you research, Observatory fits

Write in your customers’ words, not your category’s
Find the objections before the sales call does
See what you are weighed against, and why

Validate demand and language before you build
Hear how people describe the problem, in their own words
See what they use today and where it falls short

Test with prompt sets that look like your users, not your team
Build eval sets in every language your users write in
Find the follow-ups your assistant should expect

Start interviews already knowing the patterns
Category research for every client, in an afternoon
Every number shows its sample and range

What worries people before they choose a CRM?

Price after the first year comes up most.

“is the free plan actually free or do they make you pay once you grow”

Price as the team grows41%

Lock-in and moving data22%

Setup time19%

observatory · askn = 38,412

How do freelancers describe invoicing before they pay for a tool?

They describe chasing, not invoicing.

“client says they never got the invoice again. how do i send reminders without being awkward”

Getting paid on time46%

Looking professional24%

Taxes at year end18%

observatory · askn = 15,730

Build a 2,400 prompt set for first-time home buyers

2,400 prompts, tagged by intent.

{"prompt": "how much do i actually need saved for a 300k house", "intent": "budget"}
{"prompt": "is it dumb to buy now or wait for rates", "intent": "timing"}
{"prompt": "what does pre-approved actually mean", "intent": "process"}
{"prompt": "¿cuánto necesito ahorrado para comprar?", "lang": "es"}

observatory · search_promptseval.jsonl

Compare how US and UK drivers talk about renewal

US drivers ask why. UK drivers ask how to switch.

Understand why · US41%

Understand why · UK29%

Switch · US15%

Switch · UK27%

observatory · compare_segmentsn = 17,055

Give your agent the words people actually use

Early access Talk to us

Extract insights from millions of real AI conversations

Every decision today starts in AI chat Until now, nobody could see them

Turn AI conversations into decisions you can act on

Phrasing

Decision drivers

Alternatives

Friction

Personas

Trends

Prompt sets

Written the way people write In the languages they think in

However you research, Observatory fits

Give your agent the words people actually use

Get early access

You’re on the list