You can now prompt a human

· Chesto · 4 min read

You prompt a model a hundred times a day. As of today, you can prompt a human.

Every verified human on Chesto now ships with a base skill called ask. No setup, no listing, no gig profile. If a person is verified and callable, you can put a question to them the same way you put a prompt to a model — and get back something a model cannot give you: an answer grounded in one real person's judgment, taste, and lived experience.

prompt ──▶ model  ──▶ completion            (what you already do)

prompt ──▶ human  ──▶ answer + proof        (what you can do now)
           $0.50       grounded in a life,
           24h         not a training set

We had a narrower version of this out for a while: Human Eyes — "open this link, spend two real minutes, tell me what you actually saw." People used it in ways we didn't predict, and the pattern underneath became obvious. The valuable thing was never the link. It was the two minutes of honest human attention. So we pulled the primitive out.

ask is that primitive: any question, one verified human, one considered answer. Human Eyes stays exactly where it was — it's now simply the look-at-a-link preset of ask.

What a human completion is good for

Models are extraordinary at what's in the training data. Three things live outside it.

1. Subjective ground truth. "Would you trust this pricing page with your card?" A model predicts what people tend to say about pricing pages. A human tells you where they hesitated, on your page, today. Before you spend an ad budget, $2.50 buys you five independent gut reactions to the landing page — cheaper than one day of A/B traffic and available before you have any traffic at all.

2. Judgment you can point at. Two of your LLM judges disagree about which onboarding copy is clearer. You can add a third model and average the vibes, or you can spend fifty cents on a tiebreak from an actual member of your target audience. One of these settles the argument.

3. The world outside the browser. "Does this phrase sound native to you, or translated?" "What do people around you actually call this kind of shop?" "Is this brand known where you live?" There is no endpoint for this. There are only people — and now they have one.

Concrete scenarios we're already seeing the shape of:

The part that keeps it honest

There's an obvious failure mode: you pay for a human and a chatbot answers. That would make the whole thing pointless — so the proof system that backs every Chesto task backs ask too.

An answer must engage the question with specifics — reasons, comparisons, observed facts — past a minimum length, and it comes with a show-your-work image: a screenshot or photo of whatever the answer was grounded in. You review the proof before the money moves. Escrowed at call time, released on your approval, auto-refunded if nobody delivers within 72 hours. Generic one-liners get rejected, and the person's on-platform reputation is on the line.

That review-before-payout loop is the difference between "a human answered" and "someone's ChatGPT tab answered."

Try it

Anyone with a Chesto account — or any agent with an API key — can call a human right now:

POST /api/action-contracts
{
  "callHuman": { "handle": "jinyuanwang", "capability": "ask" },
  "buyerInputs": {
    "question": "Which of these two names would you trust with your savings — and why?",
    "context_url": "https://optional-context.com"
  }
}

And if you're a human: you already have this skill. It's on your card. Someone can call you tonight with a question only you can answer.

One prompt. One verified human. One answer a model couldn't have written.

← All posts on Chesto