AI for sales: 6 workflows with prompts, from research to forecast

AI for sales means using assistants such as ChatGPT, Claude or the AI built into your CRM to do the prep and admin around selling: account research, first drafts of outreach, call briefs, CRM notes and deal reviews. This guide gives six workflows with copyable prompts, shows what to check before anything reaches a customer, and lists the legal and data guardrails to set first.

How can you use AI for sales?

You use AI for sales by handing it the preparation and admin around each deal (research, first drafts, call briefs, notes, pipeline reviews) and keeping the selling itself with people. That is where the time goes: in Salesforce's 2026 State of Sales survey, the average seller spends 40% of their time selling.

The last two numbers are what sellers expect, not measured results. Still, they point at the same place this guide does: research and drafting are where AI saves the most time today.

All prompts below work in ChatGPT, Claude or Gemini. Replace the parts in square brackets. For prompts outside sales, see our list of ChatGPT prompts for work.

How do you use AI for prospect research?

AI is fastest at turning public information into a short account summary, as long as you tell it which sources to use and ask it to separate facts from guesses. Use a research or web-browsing mode, or paste the pages yourself.

You are helping me prepare to contact [Company] ([website]).Use only these sources: their website, their last two press releases,their careers page and [paste or link anything else].Give me:1. What they sell and to whom, in two sentences.2. Three things that changed in the last 6 months (hires, launches, funding, expansion), each with the source link.3. Two problems our product [one-line description] could plausibly help with, marked as "guess" and explained.4. Who probably owns this problem (job titles, not names).5. Anything that suggests they are a bad fit.If you can't find something, write "not found". Do not fill gaps.

Two habits make this reliable. First, open every link the model cites before you use a fact; models still produce confident errors, which our guide to AI hallucinations explains. Second, ask for the "bad fit" signals, because research prompts that only look for reasons to call will always find them.

For named people, stick to what they publish professionally, such as their company bio or posts. Do not ask AI to dig up personal details, and do not store guesses about individuals in your CRM.

How do you write sales outreach with AI that doesn't sound like AI?

Good AI outreach starts from one true, specific reason to write, which you supply. The model's job is to say it briefly in your voice; it should not invent the reason.

Compare two prompts:

Write a personalized cold email to the VP of Sales at Acme about our product.

Write a cold email, under 90 words, from me ([name], [role] at [company])to [title] at [prospect].The reason I'm writing: [one fact from research, e.g. "they opened an officein Munich last month and are hiring 6 SDRs there"].What we do in one line: [description].Proof: [one customer result you can verify, or leave this out].Ask: a 20-minute call next week, or a reply with the right person.Style: plain, no flattery, no "I hope this finds you well", no exclamationmarks, one short paragraph plus the ask. Give me 3 versions with differentopening lines.

The weak prompt forces the model to guess, and guessed personalization ("I loved your recent post!") is the easiest kind for buyers to spot. The better prompt gives it a fact, a limit and a style, so you get drafts worth editing. Write two or three follow-ups the same way, each adding something new rather than "bumping this to the top of your inbox."

Before anything is sent:

• Read it aloud. Cut any claim you can't back up.
• Check names, titles and company facts against the source.
• Make sure the email meets the law where the recipient is. In the US, the FTC's CAN-SPAM guide says the rules apply to business-to-business email too: no misleading headers or subject lines, a valid postal address, a clear way to opt out, and opt-outs honored within 10 business days. In the EU, GDPR and national e-privacy rules apply to cold outreach.

How do you prepare for a sales call with AI?

A call brief is the highest-value five minutes of AI use in sales: one page that reminds you who you are meeting, what happened so far and what you want from the call. Feed it your CRM history and research notes, not only the company name.

Here are my notes on [account]: [paste CRM history, last emails, research summary].Tomorrow I have a [discovery / demo / pricing] call with [titles].Write a one-page call brief:- Where the deal stands, in three bullets, with dates.- What each attendee likely cares about, based only on the notes.- My goal for this call and the one thing I need to learn.- Six open questions to ask, ordered from broad to specific.- The three objections most likely to come up, each with a short, truthful answer.- Anything in the notes that contradicts itself or is missing.

Then rehearse. Ask the model to play the buyer: "Act as the [CFO] at [account]. You are skeptical about cost and switching effort. Ask me hard questions one at a time and wait for my answer. After ten questions, tell me where my answers were weak." Practicing objections out loud with a model is cheap, private and surprisingly close to the real thing.

How do you turn sales calls into CRM notes with AI?

Paste or export the call transcript, give the model a fixed template that matches your CRM fields, and correct the result before it goes in. A fixed template keeps notes comparable across reps and deals.

Turn this call transcript into CRM notes using exactly this template.Quote the buyer where it matters. If something wasn't said, write "not discussed".Summary (3 sentences):Pain / problem in the buyer's words:Current solution and why they'd change:Decision makers and their role in the decision:Decision process and timeline:Budget signals:Competitors mentioned:Risks or blockers:Agreed next step (who, what, by when):Follow-up email to the buyer confirming the next step (under 100 words):Transcript:[paste]

Three rules keep this safe. Get consent before you record; recording laws differ by country and, in the US, by state, and some require every participant to agree. Use an AI tool your company has approved for customer data, not a personal account. And send the confirming email yourself, because it is how the buyer checks that you heard them right. Our guide to AI meeting notes and email covers transcription tools and note formats in more depth.

Many CRMs now have assistants or connectors that read records directly, which saves the copy and paste. HubSpot's connector for ChatGPT, for example, lets an admin enable access for the organization, and users only see the CRM data they are already allowed to see in HubSpot.

Can AI help with sales forecasting and deal reviews?

AI helps most by questioning the forecast, not by producing it. Give it an export of open deals and ask it to flag the ones whose data doesn't support their stage or close date; you then decide what to do.

Here is an export of my open deals with stage, amount, close date,last activity date, next step and contacts: [paste CSV].For each deal in "Commit" or "Best case":- Flag it if the next step is empty, older than 14 days, or has no date.- Flag it if there's no contact with a budget or decision role.- Flag it if the close date has moved more than once (use the history column).- Flag it if last activity is more than 21 days ago.Return a table: deal, flags, one question I should answer before the forecast call.Do not change any numbers and do not predict outcomes.

The thresholds (14 days, 21 days) are examples; use your team's own rules. Ask the model to show its working in a table so you can check it, and check any sums with a spreadsheet formula, because language models are unreliable at arithmetic over long lists.

This works only if the CRM data is decent. In the same Salesforce survey, 51% of sales leaders using AI said disconnected systems slow down their AI work, and 74% of sales professionals said they are focusing on data cleansing. A weekly 15-minute "fix your fields" block does more for AI forecasting than any tool.

Which AI tools work best for sales teams?

Most sales teams need three kinds of AI tool: a general assistant for research and writing, the AI inside the CRM for anything that touches records, and a call recorder that produces transcripts. You can run every workflow in this guide with the first kind alone and add the others when volume grows.

Pick tools by where your data already lives. If your notes, emails and deals sit in one CRM, its built-in assistant or an official connector avoids exporting data into yet another system. Check three things before you buy: whether customer data is used to train models, which region the data is stored in, and whether admins can see and switch off what each user connects. Prices and features change often, so compare them on each vendor's own pricing page when you decide.

Whatever you choose, keep the prompts from this guide in a shared document. They outlast any single tool, and a new hire gets the team's best practice on day one.

What guardrails should a sales team set for AI?

Set the guardrails before you scale AI across a team: what data may go into which tools, who reviews what before it reaches a customer, and which laws apply to automated outreach. Most problems in sales AI come from skipping one of these.

Write these into a one-page policy and train the team on it once. The policy matters most for the fastest-moving reps, who are the ones most likely to automate first.

How do you start using AI for sales this month?

Start with one workflow, measure the time saved for two weeks, then add the next. A realistic first month looks like this:

1. Week 1: call briefs. Use the call prep prompt before every meeting and note how long prep takes.
2. Week 2: CRM notes. Run every recorded call through the notes template and correct the output.
3. Week 3: research and first-touch emails. Use the research and outreach prompts for ten accounts; compare reply rates with your usual emails.
4. Week 4: pipeline review. Run the deal-flag prompt before your forecast call and fix what it finds.
5. Write down what stuck. Save the prompts that worked in a shared document, with one good example output each, and agree on the guardrails above as a team.

If your team also builds its own sales tools with AI coding agents, Orkio runs those agents on an always-on cloud box.

Frequently asked questions