AI SEO Expert Work + Which AI Models To Use For Pages, Briefs And Schema (Real GoldieBench Scores)

An AI SEO expert now hands a lot of drafting to AI models, so the practical question is which models deserve that work.

This page answers it with scores read directly from live GoldieBench pages, each quoted with the page it came from.

Then it covers the parts of the job no model does, where Julian Goldie's team comes in: book a free strategy session if you want that side handled.

What GoldieBench actually scores

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

Each result must be a single self-contained HTML file.

Every score is Julian's own 0–10 judgement from his head-to-head guides.

So the bench is a strong read on building pages and interfaces, and a weak read on brief writing or schema accuracy, which it doesn't test directly.

Use the page-building scores as a proxy and check every SEO draft by hand.

The tasks behind the keyword

Most of the drafting an AI SEO expert hands to a model falls into four buckets.

Landing Page task: the closest match

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

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

Four models tie at the top, so the tiebreaker is how each one handles your own brief.

A few tenths on one task is a small gap.

The leaderboard view, with the "Pages" tag

The GoldieBench leaderboard ranks models by average score across every task they were scored on.

• Fusion is #1 with an 8.59 average.
• Claude Opus 5 is #2 at 8.27.
• GPT-5.6 Sol is #4 at 8.16 and is tagged best at Pages.
• GLM-5.2 is #11 at 7.77, also tagged best at Pages.
• Claude Opus 5.5 is #13 at 7.57.
• Gemini 3.6 Flash is #18 at 7.08, tagged best at Pages.

For SEO drafting, the Pages tag tells you more than overall rank, since games and simulations are irrelevant to the job.

A second page task: the promo build

The AIPB Promo task is also in the Page category.

Its page lists 26 models tested, 20 scored, an average of 7.48/10, and Qwen 3.8 as the winner.

Claude Opus 5's model page shows 8.3 on that task and 8.7 on the Webos page task.

Claude Opus 5.5's model page shows 7.2 on AIPB Promo and 8.7 on Webos.

Across the page tasks, Claude Opus 5 holds steadier than Opus 5.5, which matters if you want one model for every page draft.

Picking a model pair for SEO drafting

Pick two, not one.

1. Choose one top Landing scorer, for example GPT-5.6 Sol or Fusion.
2. Choose one model with steady scores across both page tasks, for example Claude Opus 5.
3. Give both the same brief, unchanged.
4. Score each draft on three things: did it answer the customer's questions, did it invent anything, would a buyer trust it.
5. Keep the winner for that page type and retest when a new model lands on the bench.

That copies the bench's one-prompt, same-rubric idea and applies it to your business.

Where models trip up on SEO drafts

A high bench score doesn't stop the usual mistakes, so check for these in every draft.

• Invented facts. Models fill gaps with confident detail: years in business, awards, review counts. Delete anything you can't prove.
• Schema that doesn't match the page. Every JSON-LD value should appear in visible text; a phone number or address that differs is a red flag.
• Commodity copy. Without your own examples, a model writes the same page as everyone else, which Google's guide calls commodity content.
• Page sprawl. Asking a model for a page per query variation drifts into what Google's spam policies call scaled content abuse.

The fix is the same each time: feed the model your own material and edit the result by hand.

How to read a tie at the top

Four models share 9.0 on the Landing Page task, which means the bench can't separate them on that prompt.

When that happens, look at a second page task and at the model's whole scorecard for consistency.

Then let your own two-model test break the tie on your real brief.

Small-business example: an independent opticians

1. The owner lists the questions patients ask about children's eye tests.
2. Two models draft a service page from those questions and the practice's own process notes.
3. The owner deletes anything clinical the optometrist can't stand behind.
4. One model drafts LocalBusiness JSON-LD from the contact page, and every value is checked against the site.
5. Only the edited page goes live, and its AI impressions are tracked in Search Console.

What the platforms say the page still needs

A model drafts; the page must still meet each platform's bar.

Google's AI optimisation guide favours non-commodity content with a unique point of view over summaries of what's already published.

It also says Google Search doesn't use llms.txt, so don't spend a model's time generating one for Google.

OpenAI says ChatGPT search needs OAI-SearchBot access, and Bing wants entities defined consistently.

Google's AI features page adds that no special schema is needed for AI Overviews.

The human part of the job

Models can't choose priorities, earn genuine links, or decide which claims are true.

That is the AI SEO expert's real value.

Julian Goldie is a worked example. His about page lists 1,000+ SEO and link building campaigns and 240+ clients, which are his own published figures.

The GoldieBench scores are his too, with every demo built inside his Agent OS, and he shares AI SEO tests with 425K+ YouTube subscribers by his own count.

His agency, Goldie Agency, describes its link building as "manual outreach, no PBNs".

Nobody, Julian included, can guarantee a #1 ranking; Google's hiring guide says so outright.

If you want a person to set priorities for your site, book a free SEO strategy session and bring your best model drafts along.

Want to run the model tests yourself? Learn AI SEO yourself inside the AI Profit Boardroom.

Also On Our Network

• Agent Operatorsthe six-step weekly process an AI SEO expert actually runs
• AI Income Deskwhether a small business should hire an AI SEO expert, and what to ask
• AI Tool Verdictfive ways to hire an AI SEO expert, scored out of 10
• agentos.guidefive Agent OS prompts that do an AI SEO expert's groundwork

Pick two page-strong models from the bench, edit every draft yourself, and book a free strategy session or join the AI Profit Boardroom for the calls only an ai seo expert makes.