# What Picked does, screen by screen

> Every Picked feature: daily prompts on six AI models, the citations behind each answer, AI crawler analytics, articles as pull requests, client reports.

Canonical: https://picked.so/features
Last updated: 2026-11-06

Everything Picked does for AI search monitoring, as built today: the prompts it asks every day, the pages behind each answer, the AI crawlers on your site, the articles it writes for the prompts you lose, and the reports you send. Each part says which plan has it.

- **0 of 70** answers named the two products we run, on five AI models
- **91%** of the sites cited were cited by one AI model only
- **61%** of the brands named were named by one model only

LaunchOnIt and Leado, five buyer questions each, September 27 and 28, 2026, US. Details on the AI model pages.

> Tracking without writing is a smoke alarm with no extinguisher.

A dashboard that lists the prompts you're missing from and stops there hands the hard part back to you. That is why Picked doesn't stop at the number: on Growth and Scale, each prompt you lose can become a brief, an article and a pull request on your site.

## What does Picked track every day?

The prompts you choose, asked again every day, and what Picked keeps from each answer:

**Prompts.** The questions a buyer asks an AI before choosing a tool, like "best CRM for a five-person agency". Each prompt is asked again every day, because the same prompt can get a different answer tomorrow.

**Six AI models.** ChatGPT, Claude, Gemini, Perplexity, Google AI Overviews and Google AI Mode. Each plan includes some of them, and the [pricing page](https://picked.so/pricing) says how many. Claude is asked less often, the others daily.

**Mentions and position.** Whether an answer names you, and where in its list: first or fifth. The [AI rank tracker](https://picked.so/ai-rank-tracker) page explains how a place in an AI answer is counted.

**Share.** Out of all the answers to your prompts, the percentage that name you: your share of voice in AI answers. It is the number to move and to put in a report.

**Rank and sentiment.** Where you stand among every brand the answers name, like "#4 of 31 brands named", and how the answers present each one, from 0 to 100.

**Competitors.** Every brand you track gets its own page: how often it is named, for which prompts, and head to head against you. Brands the answers keep naming that you don't track yet are listed, one click from tracked.

**Demand.** Each prompt shows how many people search the closest question on Google each month in the US, so you can tell a question people ask from one nobody does.

**Topics.** Group prompts under a label like "pricing" and filter every screen to it.

**Prompt ideas.** Add prompts lists ideas from your own data: what the AI models searched while answering your prompts, and the questions behind the keywords your competitors rank for.

## What shows why a competitor won?

A number without a reason is a screenshot with a chart on it. So every answer keeps its evidence, and each prompt has its own address, so a link can stand in for a screenshot:

| Screen | What it shows |
| --- | --- |
| A prompt | The real answers, day by day, with your share by AI model and how it moved |
| Sources | The pages each answer cited, whose they are, and whether yours is one of them |
| Searches | What the model looked up before answering, when it shows them |
| Brands | Who is named instead of you, and on which prompts |
| Analytics | Which pages the AI models cite across all your prompts |

## Which AI crawlers read your site?

When someone asks ChatGPT or Claude a question, the assistant may open pages on the web to answer it. Those visits show up in your logs under names like `ChatGPT-User`. Picked reads them and splits them by why the bot came: a live answer, an AI agent, a search index or a training crawl.

You connect once per project, and which way depends on where your site runs:

| Your site runs on | Connect with |
| --- | --- |
| Cloudflare | [Connect Cloudflare](https://picked.so/integrations/cloudflare): sign in and approve, 30 days read on connect |
| Vercel | [Connect Vercel](https://picked.so/integrations/vercel): a log drain, Pro or Enterprise teams |
| Anything else | [Your server](https://picked.so/integrations/server): an open-source package for Next.js, Express and other servers, or a plain HTTP call |

The Analytics screen then shows the pages the bots opened, which bots came, and the requests they were turned away on. It also counts the people who clicked through from an AI answer, by assistant and landing page, from the referrer or ChatGPT's `utm_source` (Vercel, your server, or Cloudflare on a plan that shares referrers). "Verified only" keeps hits whose IP address matches the ranges the AI companies publish.

## How does Picked help you win a prompt you lose?

Content Gaps has three tabs:

**Content Gaps.** The prompts where the AI models name other brands and never you, with the demand for each.

**Get listed.** The roundups, alternatives pages and review sites the answers cite when they name your competitors and not you. For each one Picked finds who runs it and drafts the ask, with your entry in their format; you send it, and Picked tells you when the page names you. Getting onto one is often quicker than writing your own page.

**Keyword Gaps.** Google searches a competitor ranks for in the top 20 and you don't, with Google's page for each and the AI Overview on it. "Track as prompt" turns a keyword into the question a buyer would ask an AI.

A lost prompt or keyword has a button that starts an article. On Growth and Scale, the article goes through the same steps every time:

1. **Brief.** The searches the AI models ran for the prompt and the pages they cited instead. You read the angle, the outline and the facts to use, and say yes before anything is written.
2. **Draft and rewrite pass.** Line one answers the prompt in a sentence a model can quote whole. A second pass flags lines that read like AI and rewrites them, and you approve each rewrite.
3. **Visuals.** Sourced numbers become a chart or a table, a process becomes a diagram, and your product shows up in screenshots, all in your brand's colour.
4. **Fact-check.** A claim with no source on your site or a public page is cut, with a note saying why.
5. **Pull request.** One pull request per article on your site's repository, in the format of your existing posts, opened as a draft if you want, with the reviewers you pick. Nothing is published until someone merges it. For sites that aren't on GitHub, export it as Markdown or HTML.

Once the article is live, its page compares the prompt's answers before and after: how often they name you, and how many cite the new address.

Starter has no articles. The [pricing page](https://picked.so/pricing) has the count for Growth and Scale.

## What do you send the client or the boss?

**Reports (Scale).** A client's month as a three-page PDF in your agency's name, logo and colour, with your note on page one. Picked's name isn't on it.

**Weekly email.** A Monday summary of how your share moved, with a one-click unsubscribe.

**Unlimited seats.** Every plan includes as many teammates as you want. You pay for prompts and AI models, the things that cost us money.

## What can I see before paying?

The [free AI visibility checker](https://picked.so/free-ai-visibility-checker) asks two questions about your category on ChatGPT and Google AI Overviews and scores your brand. It needs a work email, no account and no card. Picked itself starts after you pick a plan: nothing that costs money runs before that.

## When Picked won't fit

**You want a guarantee.** Nobody can promise what a model will say tomorrow. Picked shows where you're missing, which pages the AI read instead, and helps you write a better page.

**You have no site or category yet.** AI answers recommend things that exist on the web. Build and publish first.

**Your buyers don't ask AI.** A local plumber, a walk-in trade: classic local SEO pays better.

**You want a one-off screenshot.** Ask ChatGPT yourself, it's free. Picked is for asking every day and acting on it.

## How we checked

**Our answers.** LaunchOnIt and Leado, 5 buyer questions each, ChatGPT, Gemini, Perplexity, Google AI Overviews and Google AI Mode, US, September 27 and 28, 2026.

**What is built.** the product as of September 30, 2026. Nothing on this page is planned.

**Plans and limits.** the [pricing page](https://picked.so/pricing), which reads from the same limits the product enforces.

## Frequently asked questions

### What does Picked track?

Whether ChatGPT, Claude, Gemini, Perplexity, Google AI Overviews and Google AI Mode name your brand for the prompts you choose, in what position, and which pages they cite. It asks every day, except Claude, which is asked less often.

### Does it track all six AI models on every plan?

No. Each plan includes some of them, and the [pricing page](https://picked.so/pricing) says which and how many.

### Can Picked write the articles for me?

On Growth and Scale, yes: a lost prompt becomes a brief you approve, a draft, a fact-check and a pull request on your site's repository. You review and merge it. Nothing goes live on its own.

### How does Picked see AI crawlers on my site?

You connect [Cloudflare](https://picked.so/integrations/cloudflare), [Vercel](https://picked.so/integrations/vercel) or [your own server](https://picked.so/integrations/server). Picked keeps only the AI bots' visits and shows which pages they opened.

### Is there a free version?

There is a [free AI visibility check](https://picked.so/free-ai-visibility-checker) for one site, with two questions on two AI models. The full product needs a plan.

### Does it work for agencies?

Yes. One project per client, unlimited team seats, and on Scale a white-label PDF report. See [Picked for agencies](https://picked.so/ai-visibility-for-agencies).
