01What Gemini Decision 1 Is and How It Works

Gemini Decision 1 is a new free Google model that makes decisions instead of writing answers. You give it an input and a list of questions, and it scores every allowed answer in one pass with a confidence number.

It comes with a new tool called the Gemini Decisions API. An API is just a doorway that lets your other software talk to Google's model. You can try it right now inside Google AI Studio, Google's free website for testing its models, and there is a free tier.

Normal AI models write. You ask a question and they type an answer one word at a time. That is great for an email or a blog post.

Gemini Decision 1 skips the writing. It reads what you give it, looks at the answers you allowed, and scores all of them at once.

Google says it does this in about a fifth of a second. In Julian Goldie's own tests, most answers came back in around half a second, including the trip over the internet. It can answer up to a thousand questions in one go.

02Main Features: Question Types, Inputs and Confidence

Gemini Decision 1 answers three kinds of questions:

  • Yes or no: is this email a complaint? Does this post break a brand rule?
  • Pick one from a list: is this email a refund, a where's my order, a product question or a wholesale inquiry? You can give it anywhere from 2 to 255 options.
  • A score on your own scale: score this lead from cold to book a call today.

It reads almost anything you give it:

  • Text, images, audio, video and PDFs.
  • A mix in the same request, like a recording of a sales call plus the notes from your form.

Every answer comes back with a confidence number between zero and one. So you do not just get a yes. You get a yes, 97% sure.

That one detail is what makes it safe to build real automations on. You can let sure answers act on their own and send unsure ones to a person.

03Gemini Decision 1 vs Jev: How Google Compares Them

The easy way to think about it is this: Gemini Decision 1 is Google's version of Jev. Google clearly sees it that way too. In its own launch post, Jev is the main model it compares against on Google's own charts.

Here is what those charts show:

  • Decision 1 comes out slightly ahead of Jev on most of the tests.
  • Jev still wins a couple of them.

Julian points out that these are Google's numbers, so treat them as Google's numbers. A company's own launch charts are a starting point, not the final word.

For a business owner, the big news is simpler. Most of the jobs you want to automate are decisions. Which email first, which lead first, which photo goes live, which AI gets which job.

Now Google has a fast model built just for those jobs, and you can try it for free. If you already use Jev, the best test is to run the same real questions through both and see which answers you agree with more.

04How to Set Up Gemini Decision 1 Step by Step

Here is the setup Julian walks through in this episode:

  1. Go to Google AI Studio and open the Gemini Decisions playground. Paste in a real email from your inbox, write three questions and run it. This takes about two minutes.
  2. Get a free API key from AI Studio. That key lets your other tools use the model.
  3. Open Claude Code and describe what you want in plain English. For example: build me an inbox sorter using the Gemini Decisions API with these four questions, and send anything below 80% confidence to me.
  4. Connect it to what you already use: your email, your store or your lead form.

In the playground, your input goes on one side and your questions on the other. You hit run, and each answer appears with a bar showing how confident it is.

Any tool that can send a web request can talk to the model, including the HTTP step in n8n. It also works with tools like LangChain, Vercel and OpenRouter. Claude writes the code, and Gemini Decision 1 makes the decisions.

05Old Way vs New Way: Why Decisions Matter

The old way looks like this. You open your inbox in the morning and there are 80 emails. You read every one to work out what matters.

If you do use AI for this, you paste everything into Claude or ChatGPT one at a time. That burns through your usage limits on sorting jobs. It adds up to hours a week spent deciding what to work on before you work on anything.

Here is the new way. Every email, lead, photo and post goes through Gemini Decision 1 the moment it arrives. It answers your fixed questions in half a second, and your automation acts straight away:

  • The angry customer gets flagged to your phone.
  • The where's my order email gets the tracking link.
  • The hot lead gets a booking link.
  • The broken post never reaches the client.

Only the work that actually needs writing goes to Claude. Decision 1 handles the sorting, and Claude does what it is best at, the writing and the building.

06Systems for Online Stores: Traffic Controller, Inbox and Photos

Julian built six systems for two made-up businesses. The first is an online store called Kindred Goods, selling hoodies, bottles, sneakers, backpacks and a few home items.

  • AI traffic controller: one question, which AI should do this job, with four choices. Claude for big careful builds and code, Grok Bot for live news and what people say on X, Hermes for jobs that run on their own like overnight checks, and Gemini Flash for quick tasks like summaries and translations. Eight store jobs got eight sensible calls, each in about half a second.
  • Store inbox sorter: what kind of email is it from a list of six, is the customer angry, do they threaten a bank dispute, and how urgent is it. A question about whether a jacket runs small came back as a product question, not angry.
  • Listing photo inspector: five questions per photo, including clean white background, sharpness, whether the product fills the frame, and which store category it belongs to.

The studio shots came back ready and already sorted into categories. The phone photos on a cluttered table got flagged for a reshoot, with the reason attached.

07Systems for Agencies: Lead Scorer, Brand Guard and Thumbnails

The second business is a marketing agency called Brightline Growth, running ads, SEO, content and email for clients.

  • Lead scorer: six form fills, four questions each. Is it a real inquiry or spam, does it match what the agency sells, is there a clear timeline in the next three months, and how good is the lead. A skincare brand launching a serum in six weeks scored hot. An offer of 10,000 backlinks got a near zero score and was marked spam. A request for a new website with no timeline was marked real but cold, because websites are not what Brightline sells.
  • Brand rules guard: rules like never promise medical results and always say cruelty free when talking about ingredients. One post named vitamin C and rose hip oil without saying cruelty free, and it was scored as needing small fixes.
  • Thumbnail picker: 30 real thumbnails scored in a few seconds on readability, focus, logo visibility and how likely someone is to click.

The top two thumbnails were both the evolution style, which is the style that has been winning on Julian's channel. Because it reads audio, you can also check sales call recordings for a next step.

08The Confidence Threshold and Common Doubts

The one number you must set before you let it run alone is the confidence threshold. When Julian tested an email from a customer called Jordan, the model split almost evenly between complaint and where's my order, about 46% each. That is fair, because his email was both.

If your automation just took the top answer, an angry customer threatening his bank would get a robot reply. So Julian uses one rule:

  • Above 80% confidence, the automation acts on its own.
  • Below 80%, it goes to a human.

Here are the three common doubts:

  • Can't I just use ChatGPT or Claude? For a few emails, yes. But those models write every time, which is slower and eats usage limits.
  • What if it gets something wrong? It will sometimes, like a new team member. Start with it only suggesting, and watch it for a week.
  • I'm not technical. The playground is a website you type into, and Claude Code does the building from plain English.

09Who Gemini Decision 1 Is For and What to Do This Week

Gemini Decision 1 suits store owners, agencies and creators who make the same calls over and over. For most stores, the biggest decision is sorting the inbox. For most agencies, it is sorting leads.

Here is the plan Julian gives for this week:

  1. Pick the one decision you make most every day.
  2. Write five questions about it, the ones you already ask in your head.
  3. Test them in the playground on ten real examples and check if you agree with the answers.
  4. Have Claude Code turn it into an automation with an 80% confidence rule.

Go one decision at a time. You can use the same idea for ad creatives, product images or carousel covers too. Make 30, let Decision 1 pick the three strongest, and test those.

In short, it is a new Google model that decides instead of writes, reads text, images, audio, video and PDFs, and gives every answer a confidence number. You can try it free today in Google AI Studio.