Most sellers treat ChatGPT as a vending machine, one prompt in, one draft out. This AI Central guide argues for the opposite: configure the model once, then stop starting from scratch. Four moves do it. Fill in custom instructions with your role and how you want to be answered, turn Memory on, keep every working prompt in one titled document, and run the whole thing daily across emails, content and personalization. The result is a setup, not a search box.
Reviewed 7 August 2026. Everything here comes from the AI Central guide Level Up Sales with ChatGPT, a first-party AI Central document. No outside benchmarks or vendor claims have been added to it.
The gap the guide is aiming at
Almost every seller already opens ChatGPT. Very few have ever configured it. The guide opens on exactly that split, telling readers they use ChatGPT and that it is now time to make it work for their sales strategy. Everything after that follows from the difference between using a model and setting one up.
The promise AI Central makes for the setup is narrow and checkable.
It’ll write better follow-ups, cold DMs, call recaps, objection replies.
Notice what it does not promise. Nothing about strategy, forecasting or account research. It is the four repetitive writing jobs that eat a sales day, which is also where a badly configured model does the most damage, because every draft comes back generic and then needs rewriting by hand.
Step one, tell it who you are before you ask it for anything
The first move is custom instructions. In ChatGPT, open the menu under your own name and choose Customize ChatGPT. Two fields matter: what you would like ChatGPT to know about you, and how you would like ChatGPT to respond.
The examples AI Central supplies are the actual lesson. For the first field it models an account executive at a SaaS startup selling to RevOps leaders, working mostly through LinkedIn and cold outbound. For the second, short and direct with no jargon, in the voice of a smart seller, and a tone like a trusted peer rather than a bot.
The pattern in those examples is specificity. Not a job title, a segment. Not a request for a professional tone, a named voice to avoid. The guide's verdict on the payoff is blunt.
This context = better answers, instantly.
AI Central's point is that the context loads before you type anything, so it stops being something you re-explain in every prompt. That is the mechanical reason it works. Custom instructions are not advice, they are a standing brief.
Step two, Memory turns one-off prompts into a running thread
The second move is Settings, then Personalization, then Memory on. The guide lists three consequences. It remembers your role, tone and ICP. It tracks recent conversations and previous asks. And it claims every future prompt becomes ten times faster, which is the document's own estimate rather than a measured result, so read it as directional.
The example AI Central uses to show what that actually buys you is the sharpest line in the document.
Write a follow-up like we did for Sarah yesterday
That example is short on purpose. The gain is not a smarter answer, it is a shorter request. Reference beats description. Once the model holds your recent work, pointing at a piece of it costs a sentence, while describing your requirements from scratch costs a paragraph and drops half the detail on the way.
Step three, the prompt bank is the step people skip
The third move is to open a Google Doc or Notion page and title it Sales GPT Prompt Library. That is the whole instruction, and the one most readers will scroll past, because it is the only step that produces nothing on the day you do it.
AI Central is direct about what it is for.
You’re building a repeatable, AI-powered workflow
This is the step that turns a good session into an asset. A prompt that worked once and now sits in a scrolled-away chat is luck. The same prompt in a titled document gets reused on a bad morning, handed to a new rep, and kept when you change tools. It is also the only portable part of the setup.
Step four, the daily jobs it should be doing
The guide then names the work itself, grouped in three buckets, where its prompts sit.
- Emails: a polite bump on a stalled follow-up thread, a crisp recap email built out of call notes, and a rewrite of a cold email so it leads with the buyer.
- Content: one sharp hook plus an example for a LinkedIn post idea, a post turned into a skimmable carousel outline, and an email repurposed into a poll idea.
- Personalization: a prospect's own post turned into a smart DM opener, and three tailored angles for a given ICP persona.
Eight prompts, all short, none of them clever. That is deliberate. Each assumes you are pasting in raw material you already have, a thread, call notes, a prospect's post, and asking for a transformation rather than an invention. This half only works because of the other half: once the model knows your role and tone, a prompt this thin still returns something usable.
The checklist, and the line the guide rests on
The document closes on four boxes: custom instructions filled, Memory turned on, prompt bank created, daily workflow integrated. Three of the four are one-time jobs. Only the last is a habit, and it is the one that decides whether the other three were worth doing.
AI Central's summary of the whole exercise is one sentence.
This is how you turn ChatGPT into your sales teammate, not your tool.
The distinction sounds soft and is actually operational. A tool is briefed from zero every time you pick it up. A teammate holds context between sessions. The instructions, the memory toggle and the prompt library all exist to move ChatGPT from the first category into the second.
What to do with this
If you have never opened custom instructions, do only that today, and write the two fields at the level of detail the guide models, a segment and a channel, not a job title and an adjective.
If you are already set up, audit instead. Confirm Memory is genuinely switched on rather than assumed on, and confirm your prompt bank exists as a document rather than as a belief that you would recognise a good prompt if you saw it again.
If you run a team, the prompt library is the artifact that travels. Instructions and memory are per-account and cannot be handed over, so the library and the four-box checklist are what you standardise, and new reps inherit a working setup instead of rebuilding one.
The guide's own closing advice is not to wait for inspiration, but to drop a prompt, polish the output and get on with it. Its last line calls prompts a good resource and the guide itself the engine. That is the right order. The configuration is what makes a thin prompt work on a busy day.
What should a salesperson put in ChatGPT's custom instructions?
Two things. In the field asking what ChatGPT should know about you, your role and who you sell to, at roughly the specificity of an account executive at a SaaS startup selling to RevOps leaders, plus the channels you actually work. In the second field, your voice rules: short, direct, no jargon, a trusted peer rather than a bot.
Where do I turn on ChatGPT's memory?
In Settings, under Personalization. Once Memory is running, the guide says the model remembers your role, tone and ICP and tracks recent conversations and earlier asks, which is what lets you point back at a piece of work instead of describing it again.
What actually goes in a sales prompt library?
The guide sorts its own prompts three ways, which is a reasonable starting structure: email prompts for bumps, call recaps and cold email rewrites, content prompts for hooks, carousel outlines and repurposing, and personalization prompts for DM openers and ICP angles. Title it Sales GPT Prompt Library and add prompts as they prove themselves.
Will the output still sound like me?
That is the job of the second custom instructions field, and the guide sets the voice deliberately: short, direct, no jargon, like a smart seller, with a tone like a trusted peer instead of a bot. It constrains the model's default register, which runs long and formal, and it is the field most worth rewriting until drafts read like your own sends.