Use AI to research and draft, not to scrape or auto-send. Paste in a company's own public text, ask for a short summary and one relevant opener, then check every claim yourself. Stay inside the channel rules: LinkedIn bans software that scrapes or automates activity, and Gmail requires one-click unsubscribe and spam rates below 0.3% for bulk senders.
Reading twenty company websites and writing twenty openers by hand eats an afternoon. An AI chatbot can turn that into a much shorter job. Trouble starts when people hand it tasks LinkedIn or Gmail forbid, or fire off its drafts unread. The chatbot is rarely what holds you back. The platform rules are.
Below are five steps, one prompt and the platform limits.
The five-step workflow
| Step | What you do |
|---|---|
| 1 | Choose a short target list yourself. |
| 2 | Copy the company's own words. Open their About page or a recent announcement and paste the text into your AI tool. Their own text is the safest input you can give it. |
| 3 | Run the prompt below on that pasted text. |
| 4 | Check everything. Confirm the quoted line is on their page, and delete any fact that wasn't in what you pasted. |
| 5 | Send in small batches from your own domain, with authentication set up and an unsubscribe route in every email. |
For later touches, plan a short AI-assisted follow-up sequence.
A prompt you can copy
Paste this into your AI tool, then paste the company's About page text underneath. Swap the square brackets for your details.
I will paste the About page text of [company]. In 60 words, summarise what they sell and to whom. Then write a two-sentence opener for a cold email from [your service] that refers to one specific line in the text. Quote the line you used. Do not add any fact not in the text.
The last two instructions do most of the work. Asking it to quote the line makes the opener easy to check, and banning outside facts cuts down on the made-up details AI tools are prone to. It won't stop them entirely, which is why step four exists.
What to check before anything goes out
- The quoted line appears on their page word for word.
- No numbers or names appear that weren't in the text you pasted.
- The email has an unsubscribe route and your real contact details.
- You're sending from your own domain with SPF or DKIM set up, in small batches.
Drafts all sounding samey? See why AI cold emails all sound alike.
The channel rules that set the limits
LinkedIn: research by hand, no automation
LinkedIn says it doesn't permit third-party software that scrapes or automates activity, including crawlers, bots, browser plug-ins and extensions. It also says such tools could export or scrape data and violate its User Agreement, and that a restricted account has to disable the software before it's re-enabled.
So the AI part of your LinkedIn prospecting happens off LinkedIn. You read profiles and posts yourself, copy what's relevant, and use AI to make sense of it. Any tool that promises to pull lists or send connection requests for you falls on the wrong side of that rule.
Gmail: authentication, unsubscribe and spam rates
Starting February 1, 2024, Google says all senders to Gmail accounts must meet baseline requirements. Its email sender guidelines ask you to set up SPF or DKIM for your sending domain and keep spam rates reported in Postmaster Tools below 0.3%. Marketing messages must support one-click unsubscribe and include a clearly visible unsubscribe link.
Want AI doing the heavy lifting in your marketing?
I build the systems that handle the boring 80 percent, so you get your week back. Done properly, with the human kept in.
Senders of 5,000 or more messages a day face extra requirements, including SPF, DKIM and DMARC. Google's FAQ also lists a spam rate above 0.3% as a case where delivery support or mitigations are unavailable.
Put those together and the case for volume falls apart. A big AI-generated blast that annoys people pushes your spam rate up, and once you cross that line Google won't help you fix delivery. Smaller, better-targeted batches are safer and usually read better too.
What AI won't do for you here
- It can state invented facts about a company, so every claim needs checking against a source.
- It can't legally or safely replace your own reading of LinkedIn, because scraping and automation are banned there.
- Authentication, opt-outs and complaint levels remain your job.
- This guide doesn't cover email law such as CAN-SPAM, because the official guidance couldn't be checked. If you're unsure what applies to you, ask a legal professional.
- Named vendor agents aren't reviewed here; their material lacked detail.
The chatbot drafts; deciding who's worth approaching stays yours.
Frequently asked questions
Can AI find prospects for me on LinkedIn?
Not through scraping or automation. LinkedIn does not permit third-party software that scrapes or automates activity on its site. You find and read profiles yourself, then use AI to summarise what you've copied.
Is it safe to send AI-written cold emails?
Yes, provided a human reads each draft before it goes and your mail setup meets inbox provider rules. Gmail expects an easy opt-out link and low complaint levels from anyone mailing in volume.
What should I paste into the AI tool?
The company's own public text, such as its About page or a recent announcement. Their own words are the most reliable input and make the opener easy to verify.
How do I stop AI making up facts about a prospect?
Limit what it's allowed to draw on and make it show its working. A model that must point to its evidence is easier to catch out, and you still read the result against the original web page before sending.
Do I need SPF, DKIM and DMARC for prospecting emails?
At least one of SPF or DKIM is expected of everyone mailing Gmail users. Mail in large daily volumes and you need all three in place.
How big should each batch of prospecting emails be?
Only as many as you can personally proofread and answer.


