AI Search Optimisation Agency Work: Which AI Models Build The Pages, With Real GoldieBench Landing And Webos Scores

An AI search optimisation agency now uses AI models for the drafting work behind the job, so the fair question is which models build the pages an AI search plan calls for.

Every score below I read myself on live GoldieBench pages on 11 November 2026, and each one names the page it came from.

GoldieBench has no SEO task, so this page is clear about what the numbers can and cannot tell you.

Need help with the parts no benchmark covers? Talk to my team on a free strategy session, or use the calendar: Book a free SEO strategy session.

The SEO tasks behind the keyword, and which ones a benchmark can inform

So the benchmark helps with one slice of the work: turning an approved brief into a clean page on the first attempt.

How GoldieBench scores, in one paragraph

The methodology page says every model gets the same fixed prompt, one shot, with no iteration and no best-of-several picking.

The output must be a single self-contained HTML file that runs in a browser.

Scores are Julian's own 0 to 10 ratings, and medals go to the top three per task.

On 11 November the leaderboard home page showed 1310 live demos, 50 tasks and 33 models, with 3 tasks in the Pages category.

Landing: the closest task to a service page

The Landing task page asks for a modern marketing landing page in one shot.

It showed 26 models tested, 21 of 26 scored, an average of 8.11/10, and Fugu Ultra listed as the winner.

Overall averages come from the models page and the home leaderboard on the same day.

Four models share the top Landing score, so any of them is a strong first pick for drafting a service page layout.

Claude Fable 5's 7.2 on this task sits well below its 8.10 overall, so task-level scores matter more than the headline average here.

Webos: when the page needs working parts

Some AI search pages carry tools: a quote estimator, a checklist, a comparison widget.

The Webos task page tests a desktop-style app with windows and three working apps, which is a fair proxy for that kind of build.

It showed an average of 7.98/10, with Fusion listed as winner.

• Fusion: 9.0/10 (gold)
• Grok: 9.0/10 (gold)
• GPT-5.6 Sol: 8.4/10
• Claude Fable 5: 8.0/10
• Gemini 3.6 Flash: 8.0/10
• GLM-5.2: 7.5/10

Fusion and Grok top both Landing and Webos, which makes them the safest pairing when a page mixes marketing copy with interactive parts.

Which model I would pick for each page job

• Service page layout: Fusion, Grok or GLM-5.2, all 9.0 on Landing.
• Page with an interactive tool: Fusion or Grok, both 9.0 on Webos.
• Promo or motion section: GPT-5.6 Sol scored 8.3 on the Aipbpromo task, whose page lists Qwen 3.8 as winner and a 7.48 average.
• A balanced all-rounder: GPT-5.6 Sol, 8.7 on Landing and 8.4 on Webos, and flagged "best at Pages" on the leaderboard.

Treat these as starting points for a draft, not as a verdict on SEO skill.

How to brief a 9.0 model so the page is worth citing

One-shot scores reward a model for what it builds from a single sentence.

Your brief should give it far more than that, because the facts are what AI answers quote.

• Paste the customer question the page must answer, and ask for the answer in the opening lines.
• Supply the real details: service area, process steps, named staff, opening hours.
• Ask for one topic per page, plain HTML text for every key fact, and no hidden tabs.
• Tell the model not to invent reviews, numbers or awards, and to leave a gap where a fact is missing.

Bing's guidelines favour pages whose facts stand on their own, and Google's helpful content guidance asks for people-first pages, so a good brief serves both.

What no model score can tell you about AI search

A 9.0 page can still be invisible in AI answers.

• Google says AI Overviews and AI Mode need a page that is indexed and snippet-eligible, per its AI features page; a nosnippet tag removes it as a direct input.
• ChatGPT search needs OAI-SearchBot allowed and OpenAI's IP ranges reachable.
• Copilot draws on the same crawl and index base as Bing, per the Bing Webmaster Guidelines.
• Google ranks unique, first-hand content as the biggest long-term factor, and a model cannot invent your first-hand experience.

That is the human and agency half of the job.

A worked example: a physio clinic rebuilding its treatment pages

1. Write the brief yourself from patient questions, with the physio's name and qualifications.
2. Give the brief to a 9.0 Landing model and ask for one HTML page per treatment.
3. Check every clinical claim with the physio, because health pages must be highly accurate.
4. Add LocalBusiness markup that matches the visible address and hours.
5. Confirm robots.txt allows OAI-SearchBot, PerplexityBot and Bingbot before publishing.
6. Watch AI impressions in Search Console for a month before changing the template.

Where an agency still earns its place

Models draft pages fast; they do not earn mentions, fix a CDN that blocks crawlers, or decide which training bots a business should allow.

Goldie Agency covers SEO, GEO, AEO and link building, with outreach it describes as manual outreach, no PBNs.

Self-reported figures from Julian Goldie put the team at 240+ clients and 1,000+ campaigns, alongside a 100% Upwork Job Success Score and 425K+ YouTube subscribers.

For the ChatGPT side specifically, see our page on choosing a ChatGPT SEO agency.

And no agency can guarantee a ranking or a citation: OpenAI says ChatGPT placement is not guaranteed, and Bing says the same about GEO.

Also On Our Network

• Agent Operatorsthe nine-step operator playbook with robots.txt lines, meta tags and schema
• AI Income Deskwhether hiring one is worth it for a small business, with 7 questions to ask
• AI Tool Verdicta scorecard rating five ways to get AI search optimisation done
• agentos.guidefive draft-only Agent OS prompts that do the agency groundwork first

Pick a 9.0 Landing model for the draft, keep the brief and the facts human, and book a free strategy session when you want the rest of the plan from an ai search optimisation agency.