AI Email Writing, Personalisation and AI Inboxes (2026)

#01What changed in the inbox?

Our view: these features reward the same things as good email always did (clear purpose, relevance, a sender people recognise) and punish vague openers and clickbait more quickly.

#02How do you write for AI summaries?

Test: paste your email into an AI assistant and ask for a two-line summary. If the summary misses your point, rewrite the opening.

What about prioritisation?

AI Inbox prioritises using signals like frequent contacts and inferred relationships. You cannot game that, but you can:

• Send from a consistent, recognisable sender.
• Encourage replies (questions, "reply with your biggest AP headache").
• Send mail people act on: confirmations, deadlines, useful content.
• Stop mailing people who never engage (sunset policy, lesson 7).

#03How do you use AI to write email?

Good uses

See Claude for lifecycle email copy, Gemini in Gmail and Workspace and Claude Projects for brand voice for tool-specific workflows.

A brief template

Audience: AP managers at mid-market companies, on trial day 3, have not created an approval rule.
Goal: get them to create their first approval rule.
Offer: 2-minute guide and a pre-built rule template.
Proof: none needed; product walkthrough only.
Voice: plain, practical, British English, no hype.
Constraints: under 120 words, one CTA, no claims about time saved.
Output: 3 subject lines, 1 preheader, 1 body.

Personalisation with AI

• Use first-party data: role, company size, plan, product events, content downloaded.
• Stay within your privacy notice and contracts. Check that your AI tool's data terms allow the personal data you send it, or pass only non-identifying attributes (role, industry, event) rather than names and emails.
• Personalise the substance, not just the name: different example, tactic or next step per segment.
• Avoid creepy: do not reference data the reader would not expect you to have.

#04What are the risks?

#05Worked example: Ledgerline

Ledgerline is a fictional accounts-payable software company. Figures are illustrative, not benchmarks.

Ledgerline's team uses AI to draft onboarding emails. Drafts are fast but generic, and one email claimed "save 80% of AP time", a figure nobody could source.

Walkthrough

1. Voice guide and brief template built; AI drafts start from them.
2. Claims list: approved, sourced claims only; anything else removed.
3. Review step: lifecycle manager approves every email before it goes live in the ESP.
4. Openings rewritten for AI summaries: "Your first approval rule takes 2 minutes. Here's a template."
5. Summary test: each email checked by asking an assistant for a two-line summary.
6. Personalisation: examples vary by ERP integration (from product data), no names sent to the AI tool.
7. Result: onboarding click rate and first-rule completion improve (illustrative), and no unsourced claims ship.

#06Common mistakes

• Sending AI-written email without review.
• Openers that summaries turn into nothing.
• Pasting customer data into unapproved tools.
• Personalising only the first name.
• Letting AI speed increase send frequency.

#07Try it yourself

Take your last three campaign emails and ask an AI assistant for a two-line summary of each. Where the summary misses the point, rewrite the first two sentences. Then write a brief using the template above for your next email.

#08Quick quiz

1. Who gets Gmail AI Overviews thread summaries? Answer: Everyone at no cost, rolling out from January 2026 starting in the US in English.

2. What signals does Gmail's AI Inbox use to identify VIPs? Answer: Signals like people you email frequently, contacts and relationships inferred from message content.

3. Where should the point of your email go? Answer: In the first two sentences, in plain language.

4. What must happen before AI-drafted email is sent? Answer: A person reviews facts, claims, tone and audience and approves it.

#09What should you check before the next lesson?

• Brief template and voice guide in use.
• Approved claims list.
• Human review step on every send.
• Openings pass the summary test.

Previous: Cold outbound email done compliantly · Next: Newsletters as citable content for AI search

Or return to the Email Marketing for Marketers track.

Related: Email copywriting and subject lines · Gemini in Gmail and Workspace · Claude for lifecycle email copy

#10Common questions

What did Gmail announce for AI in January 2026? On 8 January 2026 Google announced thread summaries for everyone, ask-your-inbox for AI Pro and Ultra, and AI Inbox for trusted testers, starting in the US in English. On 19 May 2026 Google said AI Inbox was expanding to AI Plus and Pro in the US, and was already on AI Ultra and in preview for Workspace Enterprise Plus.

How do AI summaries change email writing? Readers may see a model's summary rather than your preheader or opening. Put the point, the offer and any date in the first two sentences in plain language, so a summary carries your message accurately.

Can AI write and send our marketing emails automatically? AI can draft, vary and personalise, but a person should review facts, claims, tone and audience before anything is sent. Automated sending of unreviewed AI copy risks errors, compliance breaches and complaints.

Does AI personalisation need more data? Only data you are entitled to use. Personalise from first-party data such as role, plan and product activity, within your privacy notice. Do not feed personal data into tools your contracts and policies do not allow.