What a publisher is actually deciding
A pitch meeting feels like a judgement on your game. It isn't, quite. A publisher is trying to answer three narrower questions, and every part of your pitch should help them answer one:
- Is the core loop fun right now? Not "will it be fun", but is there a minute of play that already works? Publishers fund risk, but they want the risk to be in scope and production, not in whether the idea is any good.
- Can this team finish it? Evidence that you ship: past releases, a sensible milestone plan, a build that runs without a ten-step setup guide.
- Does the money make sense? A budget that matches the scope, a price point that matches the genre, and a realistic audience. Publishers model your game's return before they reply; help them do it.
For AI-native games there is a fourth question, which most studios don't answer until asked: what does it cost to serve a player for an hour, and what happens when the model misbehaves? Answer it before you're asked and you are immediately in the top tenth of pitches we see.
The build matters more than the deck
If you can send something playable, send it. A twelve-minute build that shows the core loop does more than forty slides. Some practical rules:
- Make it run in one step. A single executable or a store key. If it needs API keys, host the model yourself for the evaluation period, or bundle a scoped key that expires.
- Cut everything that isn't the hook. Skip the menus, the tutorial and the lore intro. Drop the player straight into the moment that made you want to make this game.
- Tell us what to try. A short note: "Talk to the harbourmaster about the ferry, then come back after day three." AI-native games are wide; reviewers need a path through them.
- Be honest about rough edges. List known bugs. It shows you know your game.
No build yet? Send an unedited capture of the core loop, two to three minutes, with your own voice explaining what's happening. Unedited matters: a trailer tells us you can edit video; raw capture tells us the game works.
The game pitch deck, slide by slide
Keep it to 12–15 slides. Every slide should earn its place. This is the structure we see succeed most often:
| # | Slide | What it must do |
|---|---|---|
| 1 | Title + one-liner | Name, genre, platforms, and one sentence a stranger could repeat. "A mystery where you interrogate guests in your own words, and they can lie." |
| 2 | The hook | The moment of play that sells it. A GIF or a single striking screenshot with a caption. |
| 3 | Core loop | What the player does every 30 seconds, every 10 minutes and every session. A simple diagram beats a paragraph. |
| 4 | Why AI | Why this game needs a model in the loop. What breaks if you remove it? |
| 5 | Comparable titles | Three to five games your audience already plays, with their price and rough performance. Be realistic, not aspirational. |
| 6 | Audience | Who buys this, where they hang out, and what they'll pay. |
| 7 | Art direction | Your target look, made by your artists. Mood boards are fine; generated concept art is a red flag for many publishers. |
| 8 | Systems and tech | Engine, model stack, hosting, offline behaviour, safety layer. |
| 9 | Inference economics | Tokens per player-minute, cost per player-hour, how it scales. See our cost guide. |
| 10 | Team | Who does what, what you've shipped, who's missing. |
| 11 | Roadmap | Milestones with dates and deliverables. |
| 12 | Budget and ask | Total budget, what you're asking for, what each milestone costs. |
| 13 | Status | What exists today, what's playable, any other funding or platform interest. |
Two common mistakes: market-size slides ("the global games market is worth $200B") tell a publisher nothing; and feature lists of twenty bullet points make it look like you don't know what the game is.
How to frame the AI in your game
Publishers, platforms and players are all wary of generative AI used to cut corners. Frame your use of AI around what the player experiences, not what the technology does.
- Lead with the verb. "You talk the guard into letting you through" is a mechanic. "Powered by a 70B parameter LLM" is a spec sheet.
- Show the authorship. Explain what's hand-made: the tone bible, the constraint system, the authored fallback lines, the rules the model can't break. Publishers want to see that designers are in control.
- Separate runtime AI from production AI. If you used generative tools in production (art, audio, code), say so plainly and explain human review. Mixing them up in a pitch causes confusion later, especially around store disclosure.
- Name the failure modes. Characters going off-tone, players jailbreaking them, latency spikes, cost overruns. Describe what you've built for each.
The best AI pitch we received last year never used the word "AI" until slide four. By then we already wanted to play it.
Building the budget slide
The budget slide is where many pitches lose credibility. A publisher will compare your number to dozens of similar projects. Show your working:
- Headcount × months × loaded cost is the backbone. Use real local salaries plus 25–40% for taxes, equipment and overhead.
- Add external costs: outsourced art, audio, localisation, QA, legal, and, for AI-native games, model fine-tuning and evaluation compute.
- Include a contingency of 10–15%. Publishers trust budgets with contingency more than ones without.
- Break it into milestones. Each milestone should have a deliverable a publisher can play and a price.
Keep runtime inference out of the development budget. It's a live operating cost, and many publishers (Ammighty included) handle it separately. Read how that works in our deal terms guide.
The first email and the follow-up
Most publishers take pitches through a form; use it. If you're emailing a person, keep it to five lines:
Hi Mara, we're a four-person team in Lisbon (previously shipped a game they could look up). Low Tide Courier is a delivery sim where every islander remembers what you promised them. Build (12 min, one click): link. Deck: link. We're raising $1.4M to reach launch in spring 2027. Happy to walk you through it.
Follow up once after two weeks if a publisher hasn't given a timeline, and once more after another two. Then stop. Use the time to improve the build; a meaningful update ("we've added the second island and cut inference cost by half") is the best follow-up there is.
If you get a no, ask what would change their mind. Often the answer is "come back at vertical slice", and that is a door left open.
Red flags that sink pitches
We pass on most pitches for reasons that have nothing to do with the idea. These are the patterns that most often end a read early:
- No clear player verb. If we can't say what the player does in the first five minutes after reading your one-liner, the game isn't defined yet.
- Comparables that don't match. Citing the biggest hits in the industry as comparables for a first game from a four-person team makes the rest of the numbers hard to trust.
- A budget with no milestones. "$2M to finish" without a breakdown suggests nobody has planned the production.
- A chatbot in a costume. For AI-native pitches: a character you can talk to, with nothing at stake in the conversation. Talking needs consequences.
- Unexamined costs. A free-to-play game with a frontier model behind every NPC and no cost model. It can't survive success.
- Builds that don't start. Missing dependencies, expired API keys, a crash on launch. Test your build on a clean machine before you send it.
None of these are fatal on their own, and we'd still rather see an early pitch than none. But fixing them before you send turns a "not yet" into a meeting.
Frequently asked questions
How many slides should a game pitch deck have?
Twelve to fifteen. Fewer is fine if you have a strong build. More than twenty usually means the game isn't clearly defined yet.
Should I send my pitch to many publishers at once?
Yes. It's normal to pitch several publishers in parallel. Tell each one if you have competing interest when it becomes relevant, but you don't have to name names.
Do I need an NDA before pitching?
Most publishers won't sign NDAs for first-look pitches because they see so many similar ideas. Ideas are rarely the protectable part; your build and team are. Ammighty will sign one on request before you send a build.
What stage should my game be at to pitch?
Most publishers want at least a playable prototype of the core loop. A vertical slice gives you much more negotiating power over terms and budget.
How do I pitch an AI-native game specifically?
Lead with the player experience, show the hand-authored systems around the model, and include inference cost per player-hour and how you handle misbehaving characters.