Free shipping across the UK
Same page, four different visitors - personalisation cycling through the variants in real time.
Generic websites convert worse than personalised ones. The data has been consistent for a decade: every well-targeted swap - city in the headline, currency in the price, the right CTA for the right traffic source - pushes conversion up by single or double digits. The trick has never been whetherto personalise - it's knowing which signals are reliable, which are noise, and how to do it without breaking your search visibility.
In 2026 there's a new wrinkle. AI assistants are themselves personalising the answers they give your prospects, before those prospects ever land on your site. Winning today means personalising both layers: the page the visitor lands on, and the answer the AI gave them to get there in the first place.
1. The six signals that actually work
You can personalise on dozens of things, but six signals cover almost every high-ROI use case. Pick one or two to start - the gain comes from doing them well, not from stacking ten at once.
Geolocation
Country, region, city from IP address - the easiest, most reliable signal for first-touch personalisation.
e.g. Show local pricing, currency, language and shipping options
Traffic source
Which campaign, ad, referrer or organic query brought the visitor in - tells you their intent before they say a word.
e.g. Match the headline to the ad they clicked
Returning visitor
Cookie or account-based recognition. Repeat visitors should never see the same first-time-visitor onboarding.
e.g. Skip the explainer, jump straight to the CTA
Account state
Logged in / out, plan tier, lifecycle stage. The most powerful signal you have once a visitor is known.
e.g. Hide the signup CTA from existing customers
Device & viewport
Mobile vs desktop, touch vs cursor, viewport width. Layout-level personalisation, not just content swaps.
e.g. Swap a hover effect for a tap target on mobile
Behavioural intent
Pages they've viewed, depth of engagement, time on site. The clearest signal of which problem they're trying to solve.
e.g. Promote the feature that matches their browse path
2. Three rules that prevent disasters
Most personalisation projects don't fail because the targeting is bad - they fail because someone broke a rule the first time they shipped. These three are non-negotiable:
Rule 1: Never personalise for crawlers
Both Googlebot and the AI crawlers (GPTBot, ClaudeBot, PerplexityBot, Google-Extended) should always see the canonical, non-personalised version. Otherwise you risk cloaking penalties from Google and incoherent representation in AI search.
Rule 2: Always have a fallback
Personalisation depends on signals that sometimes fail - geolocation lookups time out, cookies get blocked, edge headers go missing. Render a sensible default, then enhance.
Rule 3: Don't personalise the canonical content
Swap copy, CTAs, imagery - leave the underlying entity, headline H1 and structured data alone. AI assistants ingest the canonical version of your pages; if that version changes per-visitor your representation in their training data becomes random.
3. Where to do the swap
For headline-level personalisation that needs to be in the initial HTML (a hero with the visitor's city in it), do the swap server-side - ideally at the edge runtime, where Cloudflare, Vercel and Netlify expose geolocation and device data as request headers without an API call.
For everything else - returning-visitor swaps, source-based copy, behavioural CTAs - do it client-side after first paint. It's simpler, doesn't affect crawlers, and avoids cache-fragmentation headaches.
4. The new layer: AI Search Optimization
The biggest shift since this article first went live is that your website is no longer the first page your prospect sees. Increasingly, they ask ChatGPT, Claude, Gemini or Perplexity for a recommendation, get a personalised answer, and only land on the brands the AI named.
Those AI answers are themselves personalised - by user location, conversation context, account history - and the brands that get cited are the brands with the strongest, cleanest canonical entity signals across the sources each model trusts. That's what Generative Engine Optimization builds. Personalising your own page is one half of the modern playbook - making sure AI assistants personalise their answers to include your brand is the other half.
Every Geolify GEO package is built around exactly this: per-platform entity strength that gets you cited across all 7 major AI assistants. For the deeper plays on each model, the playbooks for ranking in ChatGPT, Claude, Gemini and Google AI Overviews break it down by platform.
If you want to amplify the entity signals AI models reward, the AI SEO Boost packages add the trust links and authority signals that compound on top of any GEO build.
Recap
Personalisation is no longer optional - and it's no longer just a CSS swap on your own page. Pick one or two high-signal segments, do them well, never personalise for crawlers, and treat your canonical page as the version ChatGPT and Perplexity will read. Then layer AI Search Optimization on top, so the AI answer that brings the prospect to your site already names you as the recommendation.
Get cited before the visitor lands
Geolify GEO packages get your brand named inside ChatGPT, Claude, Gemini, Perplexity, Grok, Copilot and Google AI Overviews. Keyword scoped, delivered in 14 days, from $499.
FAQ
What is website personalization?
Website personalization is the practice of dynamically changing what a visitor sees - copy, imagery, calls to action, pricing, recommendations - based on signals like their location, traffic source, device, intent, account state or past behaviour. Done well, it lifts conversion rates by 10-30% on the segments it targets. Done poorly, it confuses users and tanks your search rankings.
Does website personalization hurt SEO?
Only if you personalise for crawlers. Both Googlebot and the AI crawlers (GPTBot, ClaudeBot, PerplexityBot, Google-Extended) should always see the canonical version of the page - the same content every time. Personalise for human users client-side, or server-side based on signals that don't depend on the visitor being a known crawler. Get this wrong and you risk cloaking penalties from Google and incoherent representation in AI search.
How does personalization interact with GEO and AI search?
AI assistants like ChatGPT, Perplexity and Google AI Overviews increasingly personalise their own answers based on user location, conversation context and past queries. The brands that win in this new layer are the ones whose canonical content is so well-structured, entity-rich and trust-signalled that AI models confidently recommend them across many user contexts. Personalising your own site is one piece - making sure AI personalises its answers to include your brand is the bigger piece.
What signals can I personalise on?
The most reliable signals are: geolocation (city/country from IP), traffic source (which campaign or referrer brought them in), device and viewport, returning visitor (cookie or account), account state (logged in, plan tier, lifecycle stage), intent (search query they came from), and past behaviour on your site. Privacy law requires you to disclose anything that involves storing or profiling - check your jurisdiction.
Do I need a personalization platform?
For most sites, no. Edge runtime headers (Vercel, Cloudflare) give you geolocation and device for free, and a small piece of vanilla JavaScript is enough for source-based and behaviour-based swaps. Dedicated platforms like Mutiny, Optimizely and Dynamic Yield earn their keep at scale - hundreds of segments, A/B testing built in, audience data warehousing - but they're overkill for a simple 'show the visitor their city' personalisation.
How do I rank in ChatGPT and Claude for personalised queries?
The fastest path is to make sure your brand's canonical entity signals are strong enough that AI assistants confidently include you in answers across the variations they personalise on - location, intent, context. That's exactly what every Geolify GEO package builds: per-platform entity strength across ChatGPT, Claude, Gemini, Perplexity, Grok, Copilot and Google AI Overviews.