AI-personalised landing pages that don't break SEO

Client-side personalisation hurts SEO and Core Web Vitals. Server-side personalisation, gated on cookied audiences, doesn't. The architecture matters more than the tool — most personalisation vendors will sell you a JS snippet that creates more problems than it solves.

The architecture

Edge function reads audience cookie, swaps hero copy and offer based on a small content registry. Single canonical URL. Bot user-agents see a neutral default. Personalisation runs in under 30ms at the edge and never appears in HTML diff for crawlers.

What to personalise

• Hero copy and headline based on traffic source.
• Primary offer (free trial vs demo vs pricing) based on cookied audience.
• Social proof — show logos from the visitor's industry where you can detect it.
• Form length — shorter for known returning visitors, longer for cold traffic.

What not to personalise

Pricing, product specifications, legal copy. Anything that creates contractual confusion or that crawlers might index differently belongs on a separate page, not a personalised one.

The lift

5–15% on CVR for cookied returning visitors. Negligible SEO impact when configured this way. The teams that try to AI-generate hero copy on the fly usually trash both — stick to a small content registry and let the model help author it offline.