AI-assisted research, replies and follow-up — and where to stop

0:00 AI-assisted social selling

Here's a message that probably landed in someone's inbox this morning. Hi Ahmed, I was truly inspired by your insightful journey and passion for innovation. I'd love to explore synergies. Ahmed didn't reply. Nobody does. AI has made that kind of message free to produce, and worthless to receive. In this lecture you'll learn what AI is genuinely good at in social selling, six limits you must not cross, three prompts that respect those limits, and how to measure whether AI is helping you or quietly hurting your reputation.

0:39 Why now

Why does this matter now? Because AI is everywhere in sales tools. General assistants like ChatGPT, Claude, Gemini and Microsoft Copilot. And AI features built into the tools sellers already use: account and lead summaries in LinkedIn Sales Navigator, Breeze in HubSpot, Agentforce in Salesforce. That's a real opportunity. Research that took an hour can take five minutes. But it's also a trap. When everyone can generate a personalised-sounding message in a second, the only thing that stands out is a message that is genuinely true and genuinely relevant. AI can help you get there faster, or help you get lost faster.

1:23 The intern analogy

Here's a helpful way to think about it. Treat AI like a very fast, very well-read intern on their first week. They can read a hundred pages in a minute and give you a tidy summary. They can draft a letter. They can sort a messy list. But you wouldn't let them sign contracts, speak to your best customer unsupervised, or make up facts they weren't sure about. So the rule is: AI researches, drafts and summarises. A human verifies, decides and sends. Everything that follows in this lecture is that rule in practice.

2:04 Good uses

What is AI genuinely good at? Five things. Research: summarising a company's news, annual report or job adverts. Preparation: drafting discovery questions for an industry, so you can pick the best. Drafting: first versions of replies, posts and follow-ups that you then rewrite in your voice. Summarising: turning your own call notes into a clean record with next steps. And pattern-spotting: clustering a month of comments into recurring questions and objections. Notice what they share. AI works on information you can check, and a human makes the final call.

2:43 Six limits

Now the six limits. One: no fake personalisation. Don't send I loved your post unless you read it. Two: no invented facts. Models can be confidently wrong, so check every fact against a primary source. Three: protect personal data. Don't paste customer details or private messages into tools your organisation hasn't approved. Data-protection laws like the GDPR in the UK and EU, and the personal-data laws in the UAE and Saudi Arabia, still apply. Four: respect platform rules. LinkedIn prohibits scraping and unauthorised bots. Five: disclose AI where people could be misled. In the European Union, since August twenty twenty-six, people must be told when they're interacting with an AI system unless it's obvious. And six: you stay accountable for every message sent in your name.

3:38 Example 1: Mariam, Abu Dhabi

First example, the simple one. Mariam is a home décor creator in Abu Dhabi. Every Sunday she spends thirty minutes with an AI assistant. She pastes her comments from the week, with names and handles removed, and asks for the recurring questions. She picks two as next week's posts. She drafts three reply templates for the most common questions, then rewrites them so they sound like her. And she asks the assistant to check her next campaign caption for a missing disclosure or claims she hasn't backed up. What she never does is let the tool post, comment or message for her. Her replies get faster, and they stay personal.

4:26 Example 2: Tariq, KSA (illustrative)

Now a realistic business scenario, with illustrative details. Tariq sells route-planning software to logistics firms in Saudi Arabia. He used to send AI-written messages like the one about synergies. Replies were close to zero. Now he uses AI differently. He pastes a prospect's public careers page and latest press release, and asks for likely priorities with the supporting sentence for each, and anything he shouldn't assume. The tool notices three open dispatch-coordinator roles in Jeddah. Tariq checks the careers page himself. It's true. His message: I saw your team is hiring three dispatch coordinators in Jeddah. We help firms that size cut manual route planning. Would a two-page checklist be useful? No worries if not. Illustratively, his reply rate climbs because each message contains one verifiable, relevant fact.

5:22 Watch me do it: research prompt

Watch me do it. I'll run the account research prompt from the lesson. I start by telling the assistant its job: help me prepare for a first conversation. I give the company, the country and my offer in one line. Then the important instruction: use only the text I paste below. I paste the company's about page, a recent press release and two job adverts. I ask for three likely priorities, each with the sentence that supports it, five open questions, and anything I should not assume. And I add: if the text doesn't support a point, say not in sources. Here's the output. Priority one, expansion into Dammam, supported by the press release. Good. Priority two, cost pressure, supported by, not in sources. So I delete it. I keep what's evidenced, and I bring the questions, not the assumptions, to the call.

6:24 Common mistakes

Common mistakes. Letting AI write in a voice that isn't yours, full of words like synergy, journey and passionate. Trusting a confident summary without checking the source. Pasting a full customer export into a consumer chatbot. Buying an AI tool that sends connection requests and comments for you, and getting your account restricted. Running a website chatbot that pretends to be a human. And measuring only time saved. Track quality too: saves, positive replies, complaints, blocks and the dreaded, is this a bot? If speed rises while quality falls, you've automated the wrong part.

7:05 Culture and language

A quick word on culture and language, because AI can flatten it. Many assistants default to an American, informal tone. That might be fine for a startup founder in Austin. It may feel over-familiar to a senior director in Riyadh, or oddly stiff to a creator's followers in Karachi who write in Roman Urdu. So tell the tool the context: formal Arabic greeting, British English, Roman Urdu mixed with English. Then read it aloud. If it doesn't sound like something you'd say to that person's face, rewrite it. And never use AI translation for anything with legal or medical weight without a fluent human check.

7:51 Recap + try this now

Recap. AI researches, drafts and summarises; you verify, decide and send. Never fake personalisation, never use unverified facts, protect personal data, follow platform rules, disclose AI where people could be misled, and stay accountable. Your try-this-now action: run the research prompt on one real prospect or brand using only public text. Then count which claims you could verify and which you had to delete. If you want to go further, the AI for Sales Teams course covers AI across the whole sales process in depth. Next, we move to LinkedIn itself.