How to Network Using AI: Reconnect Without Sounding Fake

Why reconnecting after silence feels awkward

Reconnecting feels awkward because the silence itself has become the message. You assume the other person noticed it and drew a conclusion — they almost certainly did not. What they notice is the shape of what you send. A first contact arriving after years of nothing, carrying a request, reads as a transaction however warmly it is phrased. Most recruiters will tell you the same thing: they do not resent being asked for things, they resent being asked by someone who only appears when they want something.

The friction breaks down into three separate problems, and only one of them is a writing problem:

• You cannot remember the details. Which project, which year, what they were working on.
• You do not know what a message with no ask in it is supposed to say.
• Everything you draft sounds either too stiff or too familiar.

AI is genuinely useful for the second and third. It is useless for the first, because it does not know your history — and if you let it guess, it will invent a shared project that never happened. That distinction runs through everything below.

The Zero-Ask Reconnect Method

Everything that follows about how to network using AI sits on top of one rule, and the rule is not about prompting at all. The Zero-Ask Reconnect Method governs the first message only: it does three things, and nothing else. Anchor — name the specific shared thing. Offer — give something, whether that is a congratulation, a useful link, or just a memory. Exit — close on a warm line rather than a request.

The Exit is the part everyone gets wrong. A first message that ends “let me know if you ever fancy a coffee” has an ask in it — a soft one, but the recipient now owes you an answer. End on a statement instead and you have given them permission to reply whenever, or not at all. Counter-intuitively, that is what gets replies.

Pick your reason first

Before you open a chat window, decide which of the three legitimate reasons you are writing for. The right one depends almost entirely on how long the silence has run.

How to network using AI: the two-pass draft

Here is the workflow itself. One pass builds the message, a second pass strips the corporate varnish off it, and then you edit. Skipping the second pass is why most AI outreach reads like a press release.

Open a plain text note, not a chat window. Write down what you actually remember: where you worked together, what you worked on, roughly when, and anything you know about what they are doing now. Two or three lines is plenty. This is the part AI cannot do for you, and the part that decides whether the message lands.

Now open ChatGPT or Claude. The prompt below carries your anchor, the relationship, a hard length cap, and — most importantly — an explicit list of phrases to avoid. Telling the model what not to write does more work here than any amount of describing what you do want. Filling in the bracketed slots properly is what giving AI real context looks like in practice.

You are helping me write a short reconnection message to a former colleague. You are not writing marketing copy.

WHO THEY ARE: [name], who I worked with at [company] between [years]. We worked together on [project or team].
WHAT I KNOW NOW: [their new role / recent post / milestone — or write "nothing" if you genuinely don't know].
CHANNEL: [email / LinkedIn message]. Maximum [120] words.

RULES:
- The message must contain ZERO asks. No meeting request, no favour, no "let me know if". End on a statement, not a question.
- Reference only the facts I gave you above. Do not invent shared projects, jobs, dates or memories. If something is missing, leave a [BRACKET] for me to fill.
- Banned phrases: "hope this finds you well", "touch base", "reach out", "circle back", "synergy", "excited to connect", "I've been following your journey", "quick call".
- Write like one professional texting another they respect. Short sentences. No exclamation marks.

Give me two versions: one warmer, one more reserved.

Pass 1 output is almost always correct and slightly too pleased with itself. Pass 2 fixes that. Run it in the same conversation so the model still has the context.

Take version [A or B] and cut it hard.

- Halve the word count.
- Delete every compliment that is not about a specific, named thing they did.
- One idea per sentence. Break up anything with a comma doing the work of a full stop.
- Remove any sentence whose only job is to be polite.
- Keep the ending flat and warm. It must not ask for anything.

Then list, in one line each, anything in the draft you inferred rather than took from what I told you.

Read the draft out loud. If you would not say those words to that person’s face, change them. Delete anything on the inferred list that is not actually true. Then send it from your own account, one at a time — never as a bulk mailout.

Four prompts for four reconnection moments

Each of these slots into Pass 1 in place of the generic version. Run Pass 2 on all of them.

The career update note

For when they have posted a new job, a promotion, or a piece of work. The trick is picking one detail nobody else will mention.

[Name] just announced they are joining [company] as [role]. Here is their post, word for word: "[paste it]".

Write me a four-sentence direct message that:
- Mentions one specific detail from the post that most people commenting will skip over.
- Says, in one sentence, why that move makes sense given [what I know about their past work].
- Ends warmly with no question and no request.

Do not use "congratulations" as the first word. Do not use exclamation marks.

The useful thing you found

The strongest reconnection reason there is, because the message is genuinely about them. Works on old managers and mentors who you would otherwise never have a reason to contact. If the thing you found is a long report, refining the summary down before you send it is worth the extra minute.

I found something I want to send to [name], who runs [their area] at [company]. We worked together on [project] and they cared a lot about [topic].

Here is the thing: [paste the article, or the link plus two paragraphs].

Write a three-sentence email. Sentence one: what it is, in plain language. Sentence two: the specific reason it made me think of them. Sentence three: a warm close with no ask.

Do not summarise the whole article. Do not offer to set up a call.

The shared memory catch-up

For the genuinely long silences, where there is no milestone and nothing to send. The memory does the work.

I want to message [name]. We worked together at [company] around [year] and have not spoken since. What I remember: [two or three specific things — a project, a running joke, something they taught me].

Write a message under 90 words that opens with one of those memories, told concretely enough that they will know I actually remember it. Then one line acknowledging the gap without apologising for it. Then a warm close.

Do not apologise for being out of touch. Do not explain what I have been doing unless it connects to the memory. Zero asks.

The coffee chat ask

This one does contain an ask — which is why it is never message one. Send it after they have replied to something else, or after months of the odd genuine comment on their work.

[Name] and I have been back in contact since [what happened]. I would like to ask for 20 minutes to hear how they moved into [field].

Write a message that:
- Refers to our recent exchange in one clause, not a paragraph.
- Names the two specific things I want to ask them about.
- Proposes 20 minutes, and says I am happy to work around their diary.
- Gives them a genuine, non-guilt-inducing way to say no.

Under 120 words. No flattery about their "incredible journey".

One practical note if you are sending on LinkedIn: the note attached to a connection request is tighter than most guides claim. According to LinkedIn’s own help documentation, free accounts get five personalised invitations a month with a 200-character limit on each, and Premium removes the monthly cap. The widely repeated “300 characters” figure does not match what LinkedIn actually publishes. If you are on a free account, save those five for people you cannot already message directly.

Stop AI writing like a press release

Working out how to network using AI without producing obvious AI prose comes down to one thing: constraints beat instructions. “Write warmly and professionally” produces the exact corporate register you are trying to escape, because that is the average of everything the model has read.

Banning specific phrases works far better. This is a technique worth learning properly — the same principle behind negative constraints in prompting generally, and it is the biggest single improvement you can make to any drafting prompt.

Two upgrades once the ban list is working. First, feed the model three messages you have actually written and liked, so it copies your rhythm rather than a generic one — that is few-shot prompting, and it is the closest thing to a voice transplant. Second, once your ban list has settled, move it into custom instructions so it applies to every chat without being retyped. The same discipline is what makes difficult emails sound like you rather than like a template.

What is safe to paste about someone

Here is the question almost nobody asks: the person whose career you are pasting into a chat window never agreed to that. They are not your data to hand over.

In practice this is manageable, because a good reconnection message needs far less detail than people assume. A role, a company, a project name and a public post is enough. Their phone number, their reasons for leaving, or the confidential thing they told you in 2019 are not.

Worth knowing how the tools actually handle this, because most guides get it backwards. On OpenAI’s published plan comparison, training on your content is listed as “opt-out available” identically across every consumer tier — free and paid alike. Anthropic’s consumer plans work the same way.

The real dividing line is not free versus paid. It is a personal account versus a company-issued business or enterprise account, where training is off by default. If you are on your employer’s account, assume an administrator can see it. Our practical guide to AI privacy at work covers the settings in full.

What AI cannot do for you

Being straight about the limits matters more here than in most workflows, because the failure mode is not a bad document — it is a damaged relationship. There is a real ceiling on how to network using AI, and it arrives earlier than the tools suggest.

There is also a fairness point. Using AI to draft a message is much like asking a colleague to read your email before you send it — the history and the judgement are still yours. That defence stops working the moment the sentiment is not yours either. If you would be embarrassed for the recipient to see the prompt, do not send the message.

When your outreach goes wrong

Three things go wrong, and each traces back to a specific part of the prompt.

The last row is the one people handle worst. Silence after a reconnection attempt is almost always timing — a busy week, a full inbox — and reading it as rejection leads to a second message that is somehow more apologetic and more demanding at once.

When somebody does reply and it turns into a real conversation, that is a different skill — and if the conversation is likely to be a delicate one, preparing for it properly is worth twenty minutes.

Start with three contacts, not thirty. Write the anchors tonight, run both passes tomorrow, and send them one at a time across the week. That is all how to network using AI really amounts to: about fifteen minutes to learn and roughly three minutes per message after that — and unlike almost everything else on your list, it compounds quietly for years.