AI SEO Specialist Toolkit: Which AI Models Score Best On Page Tasks, With Live GoldieBench Scores

An AI SEO specialist now drafts briefs, schema and service pages with AI models, and GoldieBench's page-building scores are the closest public signal for which models handle that work well.

Every score below was read on a live goldiebench.com page, and each one names the page it came from.

After the model picks comes the part no benchmark covers: strategy, access, links and sign-off.

For that part, book a free SEO strategy session with Julian's team.

What GoldieBench scores, and why it fits page work

According to the methodology page, each model receives an identical prompt and gets a single attempt; nothing is retried or cherry-picked.

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

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

That makes the bench a test of shipping a working page from one instruction, which is close to how a specialist uses a model for a landing page draft.

It doesn't score keyword research, brief accuracy or schema validity, so those still need a human check.

The Pages category on the leaderboard

The GoldieBench leaderboard splits tasks into categories, and Pages holds three of the 50 tasks.

Several models carry a "best at Pages" tag there: GPT-5.6 Sol at an 8.16 overall average, GLM-5.2 at 7.77, Gemini 3.6 Flash at 7.08 and Claude Sonnet 5 at 7.01.

The overall #1 is Fusion at 8.59, tagged best at Games, and Claude Opus 5 is #2 at 8.27.

For SEO page work, the Pages tag tells you more than the overall rank.

The Web-OS task: a stress test for structured layouts

The Webos task page asks for a working desktop with windows, a dock and Notes, Paint and Terminal apps.

It lists 26 models tested, 21 of 26 scored, an average of 7.98/10 and Fusion as the winner.

A desktop UI isn't a service page, but it tests the same habits: holding a complex structure together and wiring every part so it works.

Three Page tasks, three models side by side

Model pages list each model's score on every task, so you can compare all three Page tasks at once.

GPT-5.6 Sol is the most even across the three, which is why it suits a specialist who wants one default model for page drafts.

Opus 5.5 peaks on the app-like Webos build, so it's a fit for interactive tools such as a quote calculator.

Sonnet 5's model page also lists an external reference score of 82.1% on SWE-bench Verified, a coding benchmark rather than a GoldieBench medal score, which points to scripting jobs like a crawl-rule checker.

Matching models to an AI SEO specialist's tasks

A worked example: a garage that wants a booking page

1. The specialist writes one brief: MOT and servicing page, the garage's own opening hours, three customer questions and two real reviews.
2. They send the identical brief to GPT-5.6 Sol and Claude Opus 5.5, the two most consistent Page scorers in the table above.
3. They score both drafts out of 10 on accuracy, clarity and invented claims, mirroring the bench's one-prompt approach.
4. They keep the better draft, cut anything unproven and add a first-hand detail from the mechanic.
5. They draft LocalBusiness JSON-LD separately and check each value against the contact page.

Run a ten-minute bake-off on your own brief

Bench scores are a shortlist, not a decision.

Take one real brief, with your services, area and three customer questions, and paste it unchanged into two models from the tables.

Mark each draft on three points: did it answer the questions, did it invent any facts, and would a customer trust it.

A model that scores 8.7 on a Landing build can still pad your page with claims you can't back up, so the invented-facts check matters most.

Keep the notes, because the same test is useful each time a new model lands on the leaderboard.

What the platforms say a model-drafted page still needs

Google's AI optimisation guide says AI-assisted content must still meet Search Essentials and the spam policies.

It asks for non-commodity content with a unique point of view, and says Google Search doesn't use llms.txt files.

Google's AI features page adds that there are no extra technical requirements beyond being indexed and snippet-eligible.

For ChatGPT search, OpenAI's crawler docs say OAI-SearchBot must be allowed in robots.txt.

A model can draft the page; only the specialist can confirm the page is true and reachable.

Where the specialist takes over from the model

Benchmarks rank output, not judgement.

The person in charge still has to choose the priority pages, set bot rules, earn links within Google's spam policies and define what success looks like.

Google's hiring an SEO guide says no one can guarantee a #1 ranking and asks whether AEO and GEO advice matches Google's own guidance.

Goldie Agency positions itself around four services, namely SEO, GEO, AEO and links, with outreach described on its site as "manual outreach, no PBNs".

Julian reports 425K+ YouTube subscribers and over 1,000 campaigns on his about page, and this bench is built inside his Agent OS.

To learn this model-plus-human workflow yourself, join the AI Profit Boardroom.

Measuring whether model-drafted pages earned their place

Track each new page against its own baseline in Search Console, including the generative AI performance report.

Bing's AI Performance report counts Copilot citations.

In our experience, ChatGPT visits usually show chatgpt.com as the referrer in analytics.

If a page drafted by a top-scoring model doesn't move, the brief or the proof is usually the weak link, not the model.

For a sibling read, see which models suit AEO page drafting.

Also On Our Network

• Agent Operatorsthe brief template and access ladder to run before an AI SEO specialist starts
• AI Income Deskwhether a small business should hire an AI SEO specialist, with owner examples
• AI Tool Verdictfive ways to get AI SEO done, rated out of 10 with reasons
• agentos.guidefive Agent OS prompts that build the evidence pack an AI SEO specialist needs

Pick a consistent Page scorer, test it on your own brief, then book a free SEO strategy session or learn the workflow in the AI Profit Boardroom for everything a model can't do for an AI SEO specialist.