How Picked works: track, find the gap, write, watch
From signing up to a published article in six steps: add your site, choose the prompts, pick a plan, read the first run, write the article for a prompt you lose, then watch whether the answers change. Here is what happens at each one, and what Picked does every day in between.
Picked is an AI visibility tool for SEO and marketing agencies, and B2B SaaS teams. Every day it asks ChatGPT, Claude, Gemini, Perplexity and Google's AI answers the prompts your buyers ask, records which brands and pages they name, and writes the article built to become the source they quote.
32 of 69
pages ChatGPT cited for a prompt on the 27th were cited again on the 28th
84%
of ChatGPT's links went to a brand the same answer had already named
2 of 114
pages ChatGPT cited were on Google's first page for the same question
LaunchOnIt and 
Leado, ten buyer questions, ChatGPT once a day, September 27 and 28, 2026, US. Small sample, two niches.
The steps
What happens from signing up to a published article?
- 1
Add your site
Your domain, your brand name and the competitors you care about. Picked suggests competitors and buyer questions from your site, and you edit them.
- 2
Choose the prompts
A prompt is a question a buyer asks an AI, like "best tool to monitor subreddits for buying intent". Ten good ones beat fifty vague ones. Your sales team knows them.
- 3
Pick a plan and the AI models
Nothing that costs money runs before a plan. Each plan includes some of the six AI models, and the pricing page says how many.
- 4
Read the first run
Once you have a plan, the first run starts straight away. Zero is a normal first number: it is the starting line.
- 5
Find the gap and write the article
Open a prompt you lose, read who is named and which pages the answer cited, and start an article from Content Gaps. Growth and Scale only.
- 6
Watch the answers change
The same prompts run again every day, so you see whether the AI models start naming you once the article is live.
Every day
What does Picked do each day?
For every prompt and every AI model on your plan, Picked asks the question and stores the answer. ChatGPT, Gemini, Google AI Overviews and Google AI Mode are read as the real product shows them. Perplexity is read through its API with web search. Claude is asked less often, because each of its answers costs much more to fetch.
From each answer it records:
Whether you are named, and where
- Position in the list is part of the score
Every other brand named
- Your competitors, and the ones you didn't expect
Every page cited
- The pages the model trusts for the question
The searches the model ran, when it shows them
- What it looked for before answering
| What | Why it matters |
|---|---|
| Whether you are named, and where | Position in the list is part of the score |
| Every other brand named | Your competitors, and the ones you didn't expect |
| Every page cited | The pages the model trusts for the question |
| The searches the model ran, when it shows them | What it looked for before answering |
Then it works out your share: out of all the answers, how many name you. Answers change from run to run, so the trend over weeks is the thing to read, not one day.
A ChatGPT screenshot is an anecdote, not a metric.
The gap
How do you see why a competitor wins?
Open a prompt and you get the real answers, the brands named, and the pages cited. That list of pages is the to-do list: those are the pages the model read for the question. Content Gaps lists the prompts you lose next to how many people search for the closest question on Google, so you can start with the ones that matter.
Real answers, not a mock-up
Brands named
- 1
Product HuntNamed by all four models33 - 2
BetaListNamed by all four models25 - 3
Indie HackersNamed by all four models22 - 4
Hacker NewsNamed by all four models20 - 5
UneedNamed by all four models17 - –
LaunchOnIt (You)Not named
The fix
How does an article get written?
Every article goes through the same five steps. You approve the brief, each rewrite and the pull request, and nothing is published until you do:
- 1
Brief
Picked reads the pages the AI cited and your own site, then shows an angle, an outline and the facts to use. You say yes, or you edit it.
- 2
Draft
Line one answers the prompt in a sentence a model can quote whole.
- 3
Rewrite pass
A second pass flags lines that sound like AI and rewrites them. You approve each rewrite.
- 4
Visuals and fact-check
Numbers become a chart or table, and any claim without a source on your site or a public page is cut, with a note.
- 5
Pull request
One pull request per article on your site's repository. You review it on GitHub and merge it, or export the article as Markdown or HTML if your site isn't on GitHub.
The crawlers
How does Picked know the AI read your page?
Picked can also read your site's traffic and show which AI crawlers opened which pages. It connects to Cloudflare, Vercel or your own server, and tells apart a live answer from a training crawl. It is optional: tracking and articles work without it.
When not to
When this won't help
Honest answer: sometimes this isn't worth your time.
You want a guarantee
Nobody controls what a model says. Picked shows the gap and helps you write a better page, and that is all.
You have no site or category yet
Build and publish first.
Your buyers don't ask AI
Classic local SEO pays better for a walk-in trade.
You want a one-off screenshot
Ask ChatGPT yourself, it's free.
Sources
How we checked
- Our answers: LaunchOnIt and Leado, ten buyer questions, ChatGPT once a day through its interface, US, September 27 and 28, 2026. The 32 of 69 figure counts pages cited for the same prompt on both days (nine prompts; the tenth had no links one day).
- Google's first page: the same ten questions, fetched September 29, 2026.
- How each model is read, and the schedule: the product as of September 30, 2026.
Questions
Frequently asked questions
Short answers, each one checked against the product as built or the docs we link to.
The first run starts as soon as you pick a plan, so you see where you stand the same day. Whether an article changes an answer takes longer, and nobody can promise it will. That is why Picked keeps asking every day.
The same prompt can give a different answer tomorrow, so one day proves nothing. In our own runs, more than half of the pages ChatGPT cited for a prompt were different the next day.
No. Tracking needs only your domain and your prompts. AI crawler analytics is optional and needs a connection to Cloudflare or Vercel, or a small package on your server.
A question a buyer asks an AI, like "best CRM for a five-person agency". It is the AI version of a keyword.
You do. You approve the brief, each rewrite and the pull request. Picked never publishes on its own.
The free AI visibility checker checks one site on ChatGPT and Google AI Overviews with a work email and no account.
Find out what ChatGPT and Google AI say about you
Add your domain and the prompts your buyers ask. Unlimited seats for your whole team, and no promises we can't keep.