What you'll learn
By the end of this lesson you can explain why an AI assistant gives better account research when it reads documents you supply, name the six sources to collect for each account, and ask for answers you can check.
Two ways to ask the same question
Ask an assistant "What are Northfield Logistics' priorities this year?" and you will get a fluent answer. It may be right. It may be drawn from an old article, from a different company with a similar name, or from nowhere at all. You cannot tell which from the answer, because a wrong answer reads exactly like a right one.
Now paste in the company's latest annual report and ask "List the priorities this document states for the coming year, and quote the passage for each." The answer is limited to what the document says, and every point comes with a quote you can find.
The first approach asks the assistant to remember. The second asks it to read. Reading is what these tools do well, so the work that matters happens before the prompt: choosing what to put in front of it.
Why memory fails for account research
Three reasons, all of which hit company research hard:
- It is out of date. A model's knowledge stops at some point in the past. The trigger you need happened last month.
- Detail on most companies is thin. There is a lot of text about the largest firms and very little about a mid-sized manufacturer. Where the detail is thin, a model fills the gap with something plausible.
- Names and titles change. People move. An assistant may confidently name an executive who left two years ago.
Assistants with web search reduce the first problem. They do not remove the need to check: a search can land on an outdated page or the wrong company as easily as memory can.
The six sources to collect
For a top-tier account, gather these before opening the assistant. Fifteen to twenty minutes is enough.
| # | Source | What it gives you |
|---|---|---|
| 1 | Annual report, or the "about" and strategy pages for a private company | Stated priorities, risks, named programmes |
| 2 | Latest results announcement or earnings call transcript | What leaders said recently, in their own words |
| 3 | Press releases and news page, last six months | Trigger events with dates |
| 4 | Open job posts in the function you sell to | What they are building and what they lack |
| 5 | Leadership page and the relevant leaders' public profiles | Who holds which role today |
| 6 | Your own material: CRM notes, emails, call notes | What they have already told you |
Save each as text or a PDF, with the link and the date you found it. That small habit is what makes checking possible later.
Before you paste anything, think about confidentiality. Public documents are fine. Your own notes may hold things a customer told you in confidence. Follow your company's policy on which AI tools may see which data.
Prompts second
With sources in hand, a prompt needs three instructions and very little else.
Source-first prompt (paste with your documents)
I am researching [company] before contacting [role]. I have attached or pasted these sources: [list them].
Answer only from these sources. Do not use anything else you know about the company.
1. List the priorities the company states for the next 12 months. For each, quote the passage and name the source.
2. List any changes in the last six months (people, sites, deals, programmes), with dates and the source.
3. List anything in the sources that relates to [the problem we solve]. Quote it.
4. List what the sources do not tell me that I would need to know.
If the sources do not answer a point, say "not in the sources". Do not guess.The three instructions doing the work are: answer only from the sources, quote the passage, and say when something is not there. An assistant that has been told it may say "not in the sources" has less reason to fill a gap with a guess.
Point 4 is the one people leave out. A list of what you do not know is your research plan for the next ten minutes, and your question list for the first call.
What goes in your notes
Keep two columns apart, always:
- Stated: what a source says, with the quote and the link.
- Inferred: what you or the assistant conclude from it.
"The report says they plan to open two new distribution centres in 2027" is stated. "So planning is probably under strain" is inferred. Both are useful. Only the first can go in a message as a fact. The second goes in as a question.
The check that cannot be skipped
Before anything from an assistant reaches a buyer:
- Open the source and find the quote.
- Check the date.
- Check that any person named still holds that job.
It takes about a minute per fact. One wrong detail in a first message to a senior buyer costs the account for a long time, so the minute is well spent. The last lesson of this course is a full routine for fact-checking.
Key takeaways
- Ask the assistant to read documents you supply, not to remember.
- Tell it to answer only from the sources, quote the passage and say "not in the sources".
- Check the quote, the date and the job title before anything reaches a buyer.
Exercise (30 minutes)
- Choose one top-tier account. Collect the six sources and save them with links and dates.
- Run the source-first prompt.
- Pick three facts from the answer and check each against the original.
- Write down what point 4 returned. That is your research list for tomorrow.
Why this works on named accounts
Put it into practice
15 AI Prompts for Account Research
For sellers who use Claude, ChatGPT or Perplexity to prepare for an account.
You leave with a set-up prompt, 15 research prompts in working order, a fact check that catches invented details and a sample of its output.
ChatGPT and Claude Prompts for Writing Cold Emails
For sellers who use ChatGPT or Claude to draft outreach and get emails that sound like everyone else's.
You leave with a set-up prompt, a research block, 10 prompts that draft, cut, criticise and fact-check, and one worked draft.
