GEO Agency UK Work: Which AI Models Help Draft Pages, Schema And Briefs, Using Real GoldieBench Scores

A GEO agency UK businesses hire will lean on AI models to draft landing pages, schema and briefs, so it helps to know which models build pages well and which tasks still need a person.

GEO, generative engine optimisation, is the work of being cited when AI search tools answer a question, and the models you use only cover part of it.

Every score on this page I read on live GoldieBench pages on 11 November 2026, and each one names its page.

Want a human view of your site too? Book a free strategy session with Goldie Agency.

What GoldieBench measures, and what it doesn't

The GoldieBench home page lists 1310 live demos, 50 tasks and 33 models, with every model given the same one-shot prompt and scored out of 10.

Its task categories are Games, Visuals, Sims and Pages.

There is no SEO writing task, no schema task and no citation test on the bench.

So the Pages category, and the Landing task in particular, is the closest proxy for GEO work: building a clear, well-structured page a person can then fill with first-hand content.

The Landing task scores that matter for GEO pages

The Landing task page asks for a modern marketing landing page in one shot, and shows an average of 8.11/10 across 21 scored attempts.

These are the scores I read there:

A 9.0 here means a polished page layout, not a page that will be cited.

Citations depend on what's written on the page and on what the wider web says about you.

The wider Pages picture

The Webos task page is another Pages test, with an average of 7.98/10; Fusion and Grok both scored 9.0/10 there, and GPT-5.6 Sol scored 8.4/10.

On the overall leaderboard on the home page, Fusion leads at 8.59, Claude Opus 5 is second at 8.27 and GPT-5.6 Sol is fourth at 8.16, with GPT-5.6 Sol and Gemini 3.6 Flash each tagged best at Pages.

Read together, Fusion, Grok and GPT-5.6 Sol look like steady picks for page drafts, and the Landing 9.0 group gives you several options if one is unavailable.

Matching models to the GEO tasks they can draft

Whatever model drafts it, Google's guidance is the yardstick for what the page needs.

What the platforms say the page needs

Google's guide to generative AI search says optimising for AI features is still SEO, and asks for non-commodity, people-first content.

It also says structured data isn't required for AI features and that Search doesn't use llms.txt.

Its AI features page says a page must be indexed and eligible for a snippet to appear in AI Overviews or AI Mode.

Bing's Webmaster Guidelines ask for consistent entity names and one topic per URL.

A model can lay those pages out well, but the facts, the experience and the local detail come from you.

The access checks no model score fixes

A perfect page is invisible if the AI search bots can't reach it.

OpenAI's crawler docs say sites that block OAI-SearchBot won't appear in ChatGPT search answers.

Perplexity's crawler page recommends allowing PerplexityBot to appear in its results.

Anthropic's crawler page says blocking Claude-SearchBot may reduce visibility in Claude's search results.

Check robots.txt and your CDN rules before you spend time choosing a model.

The UK details a model won't know

UK shops and restaurants hit a wall in ChatGPT's Business Hub: OpenAI's help page (Nov 2026) only lets local businesses of that kind claim a listing that already exists in the US.

For measurement, Google's generative AI performance reports include a countries view, so you can isolate UK impressions.

And no model, tool or agency can guarantee a citation: Google, Bing and OpenAI all say placement isn't guaranteed.

How to test a model on your own GEO page

1. Pick one real service page that already gets impressions in Search Console's generative AI report.
2. Give two or three high-scoring Landing models the same brief: your facts, your customer questions and your UK details.
3. Compare the drafts for structure, clarity and whether they invent anything; reject any draft that adds a fact you didn't supply.
4. Have a person add the first-hand detail, photos and examples only your business has.
5. Publish one, then watch the generative AI report and Bing's AI Performance citations for that URL over the following weeks.

That keeps the model choice tied to your own pages rather than to a leaderboard alone.

Model drafting gotchas for GEO pages

• Models love generic copy; Google's guide calls that commodity content and asks for a unique point of view instead.
• Models may invent statistics or reviews, which you must strip out before anything goes live.
• A model-written JSON-LD block can drift from the visible page text, which Google says it should match.
• A model can't check robots.txt on your live server unless you give it the file, so run the access check with an agent or by hand; the AI Profit Boardroom trainings cover both routes.

Where the human and agency part comes in

Models cover drafting; people cover strategy, proof, outreach and judgement.

Julian Goldie's agency, Goldie Agency, covers SEO, GEO, AEO and link building with manual outreach and no PBNs, serving UK clients remotely.

Self-reported figures on Julian's about page put the campaign count at 1,000+ and his YouTube audience at 425K+ subscribers.

For the human part of your plan, Book a free SEO strategy session.

To run the model side yourself, learn AI SEO yourself in the AI Profit Boardroom.

Also On Our Network

• Agent Operatorsthe eight-step operator process for running a UK GEO engagement
• AI Income Deskwhether a GEO agency is worth it for a small UK business, with worked examples
• AI Tool Verdictfive routes to GEO in the UK rated out of 10 with a verdict
• agentos.guidefive copy-paste Agent OS prompts to do the GEO groundwork before hiring

Pick a page-build model from the Landing scores, check your crawler access, and book a free strategy session to plan the human side with a GEO agency UK firms can trust.