Key takeaways
- Real personalisation is one specific, relevant observation tied to a reason to talk.
- An assistant needs two inputs: facts about the prospect and facts about your own offer and voice.
- Match the depth of research to the value of the account.
- Check facts and sources before anything is sent; a wrong detail is worse than none.
What information does an AI assistant use to personalise a message
An AI assistant uses two sets of information: what is public about the prospect and what you have told it about your own company. The first gives it something specific to say; the second makes sure the message connects that detail to what you actually sell.
- The company's own website: products, customers, open roles and news.
- The person's role, team and public posts.
- Recent events: a launch, a new market, a funding announcement or a senior hire.
- Your offer, the problems you solve, your proof points and your writing voice.
What makes AI personalisation feel real rather than generic
Personalisation feels real when the detail explains why you are writing now. "I saw you are hiring three field engineers in Leeds" works if your product helps field teams; it is noise if it does not.
Take an illustrative logistics software seller writing to Hartley Freight, which has just opened a new depot. A weak draft praises the announcement. A strong one says that a second depot usually means scheduling across two sites, and asks how they plan to handle it. The assistant found the fact; the angle comes from knowing your product.
Avoid details that feel like surveillance, such as personal life, family or anything not shared in a work context.
How do you keep quality high when volume grows
Keep quality high by tiering the work instead of applying one level of effort to every prospect. Your most valuable accounts get deep research and a person editing every message. A middle tier gets an assistant draft that someone reviews before sending. The long tail gets messages personalised to the segment, such as industry or role, rather than to the individual.
Review a sample every week, not only the first batch. Drafts drift as lists change, and a quick read of ten messages catches repeated openings, stale facts and claims you would not make.
Where does personalisation at scale go wrong
Personalisation at scale goes wrong when the research is wrong, out of date or irrelevant. An assistant may mention a product the company has retired, or a post from two years ago, and the prospect notices at once.
Ask for sources with every fact the draft relies on, and check them for your top accounts. Watch for sameness too: if every message opens with "I noticed", readers spot the pattern even when the details differ.
Where Pebbles Ai fits
- FoundationCompany IntelCompany Intel holds your business knowledge and brand voice, built from your website, company information and files, so every assistant writes from the same facts about your offer.
- CapabilityOutreach contentBuild email and LinkedIn campaigns of up to seven touches with a personality for each piece; generated emails are saved as drafts in your connected Gmail or Outlook mailbox for you to review.
- CapabilityResearch & analysisAsk an assistant to research a prospect's company from its website and public sources, with numbered citations you can check before you write.
Common questions
Can AI personalise outreach without sounding robotic?
Yes, if it writes from a specific, relevant fact and from a clear description of your voice. Most robotic drafts come from thin inputs, not from the model itself.
How much personalisation does a cold email need?
One sentence that shows you know why this person might care, tied to the problem you solve. More detail rarely helps and often reads as padding.
Should a person review every AI-written message?
Review every message for your most valuable accounts and a regular sample for the rest. Loosen the review only once samples are consistently right.
About this page. Written by the Pebbles Ai team for people who run sales, marketing and growth. Last updated .
