Claude for sales prospecting: research one account, then write to one person

Use Claude for the reading. Give it one account, let it search the company's own pages and its last earnings call, and ask for a research note in which every fact carries its link, its date and the sentence it came from, with anything it could not confirm listed apart. Keep what you sell and your rules in a project, so every chat starts from them. Then draft the email from the note, and check each fact before it is sent.

The reading is where the time goes. In Outreach's 2025 survey of 500 sales professionals, 49% called account research the hardest part of email prospecting, and 45% already use AI for it.

Set up a project once

A Claude project is a workspace with its own chats, its own instructions and its own knowledge files (Anthropic). Write the instructions once: what you sell and the problem it fixes, who you sell to and who you never write to, and the rules every email follows. Upload two emails of yours that got replies, as examples of your voice. Every chat in the project starts from all of it, so each chat can hold one account. Free accounts get up to five projects.

Ask for the note, not the email

An email is the last step. Ask for the email first and Claude has to fill it with something; ask for the note first and you can see what it found, and what it did not, before a word is written.

Research [company] for a first email to [name], [role].

Search the web. Read the company's newsroom, its careers page, and its most recent earnings call or annual report.

Return a research note:
- Three facts that bear on [the problem we solve]. For each: the link, the date it was published, and the sentence it comes from, quoted.
- Why now: one sentence on what changed in the last 90 days.
- Could not confirm: what you looked for and did not find. Do not guess.

Use only pages you opened. Mark any fact older than twelve months.

The quoted sentence is what makes the note quick to check: open the link, find the sentence, and the fact either stands or goes. The "could not confirm" line is what stops a guess from reaching the email as a fact. Search is on every Claude plan (on Team and Enterprise, an owner switches it on first), and answers that use it cite their sources (Anthropic). On paid plans, Research searches more widely, including connected sources such as Gmail (Anthropic); for one account, ordinary search is usually enough.

Hand it the long documents

The best facts are often in documents nobody reads: the annual report, the earnings call transcript, the investor deck. Upload the PDF and ask for the passages that bear on your problem, each with its page number and the sentence quoted, so you can find it again. Claude reads PDFs of up to 1,000 pages, but it reads the charts and images only in PDFs of 100 pages or fewer (Anthropic). For a long annual report, upload the section with the numbers you need.

Then the draft

In the same chat, ask for the email from the note: 60 to 100 words, opening on one fact, naming the problem it suggests, one sentence on what you would do, and one question to end. Ask for every sentence to be listed with the fact it uses, so a sentence with no source stands out. The full drafting prompt, with the openers to ban, is in how to write cold emails with ChatGPT, and it works as written in Claude.

Check before anything is sent

A citation is where the check starts. Open each link, find the quoted sentence, and confirm the date. Anything the note marks as old or unconfirmed stays out of the email. Then send the draft to yourself with the merge fields empty: the free mail-merge preview shows any {first_name} that would arrive as written.

Your prospects' data

On Claude's Free, Pro and Max plans, Anthropic trains its models on your chats only if you allow it. The setting is under Settings, Privacy, Help improve our AI models (Anthropic). With it on, chats can be kept in de-identified form for up to five years. There are exceptions either way: a chat you rate with a thumb can be kept for up to five years and used for training, and a chat Anthropic's safety systems flag can be used to improve its safety models. Team and Enterprise chats are not used for training by default.

Anthropic does not train on what Claude reads through the Gmail, Drive and Calendar connectors, though with the setting on, Claude's replies that repeat it, and anything you paste from those apps, can be used (Anthropic).

Whatever the plan, keep the note to what the person and their company published about the job, and leave out anything personal that is not.

Questions

Can Claude send the emails too?

It can. With the Gmail connector, Claude can send, reply to and forward email, and by default it asks for your approval before each one (Anthropic). We would keep it to drafts: the email goes out under a person's name, so a person should read it last and press send. Anthropic's usage policy also rules out using Claude to generate or distribute spam.

Is Claude better than ChatGPT for prospecting?

For this work they do the same jobs: both search the web with sources, both hold your rules in a project, and both can send from a connected Gmail account (ChatGPT on Plus, Pro, Business and Enterprise, on the web). What you give them decides the email. The ChatGPT side is in how to write cold emails with ChatGPT.

How many accounts can one person research this way?

As many as they can check. The reading is quick; the check is the limit, and it is the part that keeps a wrong fact out of an email to someone who would know. It is one reason a team of two or three runs 50 to 200 accounts (how to build the list), not thousands.

If you want this run for your accounts, see how we run it.