Your fantasy football season is an interesting theater for AI. There is a research and strategy element. There is a real-time data element. Beyond the draft, there is a season-long team management process. AI can play a role in all of these phases — from scout, to draft planning and execution and on to roster management and perhaps even picking your starting lineups.
I have written about this before. In 2024, Perplexity was helpful but ChatGPT ran into a wall and kept giving advice from last season. In 2025, deep research was on the scene but Perplexity’s real-time data was the critical advantage. The most critical failings were the limits of context windows — still more like 250k tokens at that point — and a lack of real-time access to your fantasy roster. It was great at evaluating incoming trades, but it could not scan rosters to find the targets for you. Automating any of this was limited — what access it could have was through a real-time browser connection like Perplexity’s Comet but that generally required being in front of your PC.
There have been two huge shifts in AI in the last year. The first is the rise of the agentic tools. You were still dealing with chatbots in 2025. Now you can have much meatier, longer-running projects. They can handle multi-agent, adversarial workflows. You can have something check your work. Beyond being able to hold a thought, the agentic tools are vastly more capable. ChatGPT now has a built-in browser. Claude in Chrome is amazing. Your AI operator can login to your league, read the rosters and tell you which way is up. You don’t need to paste in screen grabs to communicate what is happening.
The models have become dramatically better. The typical model today has a million token context window. You are not constrained from that perspective in the ways you once were. You can have extremely long-running threads about your draft or your team. Just like many other things, your AI tool can now be a much more useful coworker and partner.
How to Put it Together in 2026
Given the new capabilities to integrate data and use tools directly, you want to operate on a project per league basis in 2026. You can run differentiated strategies per league, just like the real world. ChatGPT Work or Claude Cowork have the capabilities required to connect to your league, read the data and give real, pointed advice on how to manage your team.
These strategies do require a paid version of Claude or ChatGPT. The free accounts don’t come with enough tooling; you need the tokens and the access to better models. Web browser access, included on paid tiers, is a critical component here. That enables a direct connection to your league. ChatGPT has a native cloud browser. Claude can use Claude in Chrome to get to your fantasy data. At the time of writing, none of the leagues make an MCP server or other sort of connector easily available. They have no incentive to — they are paid for ads, not to be a headless data platform.
I wrote this just as the new models — Fable 5.1 and GPT-6 Astra — launched. I did not use them in creating these tools. They are amazing, but I’m not sure they would add anything to what is more of a memory, context and tool access challenge.
Step 1: Defining the Problem
The first thing you need to do is get the data as well as your guidance into the system. The way to do this is to fire up a new project and start with some custom instructions. This does not need to be complicated. The tools need to know where to find the league as well as how you think about the problem. We don’t need to over instruct, just give it the parts it needs to know and to let it go find the rest. For example, this is what I use for custom instructions to get started. I am proud to say I have never played a Cowboy in fantasy save one time when Amari Cooper got traded and that was not my fault.
League: LEAGUE_URL_GOES_HERE
Team: TEAM_URL_GOES_HERE
## Standing Draft Rules
- **Ignore bye weeks.** Do not downgrade players, alter roster construction, or break ties based on bye-week overlap. Bye-week problems can be managed in season.
- **Never draft a Dallas Cowboy.** Do not recommend any player currently on the Dallas Cowboys, the Dallas Cowboys D/ST, or a Dallas kicker, regardless of value, ranking, positional need, or draft position. Only Wyatt may explicitly override this rule.
- **Roster no more than one tight end.** Draft an elite tight end early only when the value is compelling. Otherwise, select one later and stream or replace the position during the season. Do not draft or hold a second tight end for depth or bye-week coverage.
- **Roster no more than one D/ST.** A preferred defense may be drafted before the final round when the value and remaining board justify it. Stream or replace the position during the season rather than carrying a second defense.
- **Protect bench value.** Use bench spots on a third quarterback and upside running backs or wide receivers—not backup tight ends or defenses.
- **Apply these as hard constraints.** All rankings, target lists, draft recommendations, and roster advice must follow these rules unless Wyatt explicitly changes them.
- **Prioritize bankable opportunity.** Start with carries, targets, routes, red-zone work, and depth-chart security. When values are close, prefer the durable workhorse or indispensable receiver over the smaller, injury-prone, committee-dependent, or purely explosive player.
- **Draft exactly one kicker.** Normally use the final-round pick unless Yahoo roster mechanics require selecting one earlier.
- **Verify live player context.** Before current recommendations, refresh injuries, suspensions, transactions, and depth charts. Do not treat old ADP, an old team assignment, or a previous depth-chart position as current fact.
- **Never Apply Any Changes!!!** Only the human submits roster moves!
These are global rules — you don’t need to get into things like how you never, ever trust Christian McCaffrey after a few injury plagued seasons. We will get to that later.
When you fire this off you will have to work with your tool to get it logged into the league. You will have to say yes to a few scary security prompts. Claude will use chrome so you can just have a logged in browser ready. ChatGPT has a built-in cloud browser that is a bit slicker from an experience perspective, but takes a minute to get started. Either way, after a few minutes and a little finagling you should get to the point where the AI has a sense of what your strategy is and what you want to do.
Step 2: League Research
Now that you have connectivity, you can ask your AI to do your research. I like to start simple — “read up on the league and the rules” gives it license to go get the basics. It should reflect back broadly correct facts on the ground. Once the basics are in place, the next task is to dig into into history. Here is how I would build that up:
Start by asking your partner “Look at last year. How did I draft? How did that work? What lessons should I take?”
Check the scoring rules — “Do scoring rules create any unique advantages or disadvantages? How should I think about roster construction?”
Figure out your opponents — “Study my opponents’ strategies. What were they doing? How were they constructing their roster?”
Finally, save your work. Ask your AI to put it all together into a final report, and review it carefully. Once you are happy, get the AI to give you a copy of the file and add it to your project for future reference.
Step 3: Embedding Your Strategy
Now that you have learned about the league, it is time to share what you think. This is where you share things beyond those basics from the project home.
This is also a time to share — the more you can feed the research engine the better decisions it makes. Some examples of things to share with your AI are guidance on your strategy and data you use. If you live by a certain draft cheat sheet, this is the time to hand it over. You want to train this AI to think like you. Some examples and other ways to start:
Have it look at your previous year’s strategy and quiz you on it. This creates an amazing baseline for it to work with.
This is the place to share your favorite fantasy guides. I would start with the The FF Newsletter if you need something short, sweet and effective. Your cousin’s tried and true spreadsheet works here too. AI eats anything.
What players do you hate? Are there people you want to avoid? This is the place for the don’t draft list you keep in your head. Aside from Cowboys that is.
Once you have shared enough instruction, it is time to have the tools create your draft day guidance document. This is what you can use to help your AI partner guide your strategy, or just to keep you honest when you are on the clock. To get it going request that the tool do something like:
Take everything it knows about the league and your strategy and develop a comprehensive draft guide. Build a table of the team by position listing when we should be looking to draft for what. Create a round by round list of suggestions. List names to target by round in a subsequent section. Be thorough and complete.
Take this document, review and tweak it as necessary and also save it to your project’s context.
Step 4: Testing
You can’t quite trust the AI unless you evaluate it.
Start a new chat in your project. Ask the AI to help you simulate a draft. Tell it to simulate the other picks and instruct it not to check the league site for results, but to feel free to use the internet in other ways.
What you should see is that it picks somewhat like you would expect your fellow managers to pick — remember the research about the teams and the draft? It should take into account current situations. The advice should feel right. You should have a partner that can do things like advise about who is on the board for a pick, what might happen between them and help you game it out.
If it does not feel right, revisit the first few steps. What didn’t you tell it? Where is it failing? If you are not getting the right advice for the kinds of players you want, ask for corrections and apply them to the strategy file. If we are having problems with player status, upload a good, current fantasy football player guide that takes care of that and gives the AI a reliable source.
Step 5: Profit???
Once your mocking goes well you are ready for the draft. Start a clean thread, tell your tool it is draft night and let it help you orchestrate the evening. I’ve tried this in anger, here is what I learned:
Overall my AI worked — I got a good draft, it helped keep me on target and kept me thinking about the right players and positions. I think I got a better team than going alone, especially without much time for preparation. The AI was much better than winging it.
For draft night, having the AI read the website and process it is a bit too slow, especially on a 60 second draft clock. Copying and pasting from chat is easy, accurate and effective though and worked very well for me.
The draft in question was particularly long — very deep benches. The tool collapsed near the end. If I had a 16+ round draft, I would be prepared to jump to two conversations once things start to bog down. In hindsight it was obvious I had filled context, but I didn’t see it at the time.
The coolest thing I could do that I was not thinking about is that I could use it to track other managers’ strategies in near real time. I could start to game out future picks with a surprising degree of accuracy, helping me stay ahead of position runs at times. There are definitely superpowers here.
This has been a really interesting project. Draft day is just the beginning of the season, I want to see how the project develops as the season continues. Stay tuned for an upcoming check in. Good luck in your drafts!
This post and process has been about fantasy football and how to manage your team. But the basic technique described here is a universal play. Start with some rules, research the universe, add your data and build a decisioning engine. It works across the board. Fantasy football is a great way to practice something that will make you a much more effective AI practitioner and professional.
Did you enjoy reading this post? Hit the ♥ button above or below because it helps more people discover great Substacks like this one and it helps train your algorithm to get you more posts you like. Please share here or in your networks to help us grow!