Which tool to choose
The chain works best when the two AIs have different characters. One more inclined to extended, coherent writing does the draft well; one more terse and direct does the critical review well. You can also use the same AI twice with two opposite roles, but two different tools give a real cross-check: the second one doesn't defend what the first wrote, because it didn't write it.
For those who want to start without complicating things: pick the two free assistants you already use and keep two tabs open. The chain is a method, not a program to install.
How to do it
Chaining two AIs works better for a precise reason: each AI does a single task and does it carefully, instead of doing three in the same answer and losing one along the way.
- Decide the two roles before starting: who produces and who checks, or who translates and who adapts, who gathers and who synthesizes.
- Give the first AI a single, clear task. No "write and correct and shorten": just "write."
- Copy its answer in full.
- Open it in the second AI with the new task, declaring that the text is by another author to be improved, not defended.
- If needed, do a third pass: bring the corrections back to the first one for the final draft.
The operational syntax for the second AI (review):
The text below is the draft of another author. Your task is only to review it as a strict editor: find the weak sentences, the unclear passages and the points without proof. List the corrections, then rewrite the improved version.
[paste the first AI's answer here]
Between one pass and the next, read the result: the chain improves only if you check that each link does its job. If the second AI makes the text worse, the link should be redone, not accumulated.
A concrete example
Davide has to write the introduction to his business for his website. He asks the first AI: "write a warm, personal introduction to my artisan carpentry, based on these notes." He gets a flowing but somewhat generic text. He copies it into the second AI: "you're an editor: this text is by someone else, cut the clichés and make every statement concrete." The second one removes three clichés and has him add a real detail about the wood he uses. The final result is better than what either AI would have given on its own.
When it does NOT work (and how to fix it)
If the second AI rewrites everything from scratch instead of improving it
You gave it too much freedom. Narrow the task: "don't rewrite from scratch, correct only the weak sentences and explain every change." A narrow task produces a review, a broad task produces a new text that loses what was working in the first.
If the chain is slower than a single chat and isn't worth it
For a simple task, indeed, it isn't worth it: the chain serves when the work has different phases (create and then judge, gather and then synthesize). If you're doing a single thing, a single AI with a good prompt is faster. Use the chain where the value lies precisely in the change of perspective.
If the first one's output is too long to paste into the second
Reduce before passing: ask the first AI to also hand you a concise version of its result, and pass that to the second. The intermediate link that shortens is often what makes long chains work.
A tip from someone who actually uses it
The jump in quality doesn't come from the number of AIs, it comes from the fact that the second one doesn't know it's talking to the first. Always declare to it that the text is by "another author": this little fiction makes it genuinely critical, because it removes the instinct to defend the work. It's the same reason a human editor corrects other people's texts better than their own.
Frequently asked questions
Do I have to pay two subscriptions to do the chain?
No. The free versions of two different assistants are enough to try the method. You only pay if you make very long chains or work on huge texts every day; for normal use, two browser tabs are all you need.
Can I use the same AI twice instead of two different ones?
Yes, and it often works: open a new chat for the second role, so it doesn't carry over the context of the first and gives you a fresher judgment. Two different AIs only add one more point of view.
Do more AIs in a chain always mean a better result?
No, and it's the misconception to avoid. Beyond two or three links the noise accumulates and every pass introduces small drifts. The chain wins when each link has a clear purpose, not when it's long. Three targeted passes beat ten vague ones.