Welcome, let's start with an honest picture
Hi, and welcome to level one! Before we write a single prompt, I want you to have a clear, honest view of what AI assistants can do in a sales job. Tools like ChatGPT, Gemini, Claude and Microsoft 365 Copilot can save you real time. They can also land you in trouble if you trust them too much.
So here's the plan. First we'll look at what these tools actually are. Then we'll sort your everyday tasks into three buckets. And then we'll check what the evidence says about how sales teams use them, and what that evidence doesn't prove.
So what is a large language model, really?
The assistants we just named are built on large language models, or LLMs. The core idea is simpler than it sounds.
An LLM is trained on huge amounts of text to predict the next word. Give it "Thanks for your time on the call today, I've attached the..." and it's very good at guessing what comes next. Do that billions of times and you get something that writes fluent emails, summaries and plans.
Think of a new rep who has read every sales email ever written but has never met your customer. They know how good outreach sounds. They don't know whether your prospect really opened a new office last month.
That's the catch. Fluent writing and true facts are two different things.
Why can it sound so sure and still be wrong?
That's the big question, isn't it? When an AI invents a fact and states it confidently, we call it a hallucination.
In September 2025, OpenAI published research explaining one reason why it happens. During training, a model learns patterns from text, but nothing marks each statement as true or false. Patterns like spelling are easy to learn. One-off facts, like a particular company's headcount, are not.
The researchers also found that the usual way of testing models rewards guessing over saying "I don't know." It's like a multiple-choice exam with no penalty for wrong answers. If you leave a question blank you score zero, so you may as well guess.
What does that mean for you? An AI will happily give you a prospect's revenue, a competitor's pricing or a quote from a CEO, all in the same confident tone, whether it's right or not. How confident it sounds tells you nothing about whether it's accurate.
Three buckets for your everyday tasks
Let's make this practical. I find it helps to sort sales tasks into three buckets.
Bucket one: AI can help. These are low-risk jobs where a rough answer is still useful. Brainstorming questions to ask on a discovery call. Suggesting angles for a cold email. Explaining an industry term you've never heard. Playing a tough buyer so you can practise handling objections.
Bucket two: AI drafts, you check. This is where most of the time savings are. A first draft of a follow-up email. A summary of your call notes. An outline for a proposal. A tidy-up of messy CRM notes. Here's the rule: anything that goes to a customer or into a record gets checked by you first. Every name, number, date and promise.
Bucket three: stays human. Agreeing a price or a discount. Making commitments about delivery or results. Deciding whether a deal is real or wishful thinking. Reading the room when a buyer goes quiet. Building trust over months. The AI can help you prepare for these moments, but the decision and the relationship are yours.
Here's a quick test. Say the AI drafts a follow-up that mentions a 15% discount you never discussed. Which bucket was that? Bucket two, and it shows why checking matters. If that email went out unchecked, you've made a promise your manager never signed off.
What does the evidence say?
You've probably heard that "everyone in sales is using AI now." Let's look at some dated numbers.
Salesforce's State of Sales report, published in February 2026, surveyed 4,050 sales professionals in 22 countries, the UK among them, in August and September 2025. It found that 87% of sales organisations use some form of AI for tasks like prospecting, forecasting, lead scoring or drafting emails. Sellers also expected AI agents to cut prospect research time by 34% and email drafting time by 36%.
For comparison, the previous edition (July 2024, surveying 5,500 sales professionals in spring 2024) found 81% of sales teams were experimenting with AI or had fully put it in place.
HubSpot's 2024 State of AI in Sales research, based on more than 600 sales professionals, found 43% said they used AI at work.
What the numbers don't show
Now, here's where your judgement comes in. Why might those figures differ so much?
Partly because they measure different things. "The organisation uses some form of AI" is not the same as "this rep uses AI at work." A company can count as an AI user because one tool has a built-in feature most reps ignore.
Also, both are surveys run by companies that sell sales software. That doesn't make them wrong, but it's worth knowing. The answers are self-reported. And that 34% figure is what sellers expect to save, not a measured saving.
So what's a fair conclusion? AI use in sales is common and growing. These surveys don't prove that AI will make you sell more. That depends on how well you use it, which is exactly what this course is for.
Later on we'll cover approved tools, what must never be pasted into one, and how to check output properly. For now, the three buckets are enough.
Key takeaways
- An LLM predicts likely next words. It writes fluently but has no built-in way to know what's true.
- Hallucinations happen partly because training rewards guessing over admitting uncertainty, so confidence doesn't mean accuracy.
- Sort tasks into three buckets: AI can help, AI drafts and you check, and stays human.
- Salesforce's State of Sales report (Feb 2026, surveyed Aug–Sep 2025) found 87% of sales organisations use some form of AI.
- Treat survey figures carefully: note the date, who ran the survey, what exactly was measured, and whether a figure is a forecast or a result.