Generative AI SEO Services + Which AI Models Do The Drafting, With Live GoldieBench Landing And Promo Scores

Generative AI SEO services hand some jobs to AI models and keep others with people, and GoldieBench scores can only guide the model side.

Of the seven jobs, two involve a model producing a finished asset: service pages and promotional explainers.

The other five, from crawler access to citation reporting, depend on settings, relationships and judgement that no one-shot benchmark measures.

Every score below was read on live goldiebench.com pages on 11 November 2026, with the page named. To skip straight to the strategy, get a free SEO strategy session or use the calendar: Book a free SEO strategy session.

Model jobs and human jobs inside the seven services

The mapping is editorial; the bench itself doesn't label tasks by SEO job.

What the bench does and doesn't test

The methodology page says each model gets one fixed prompt, one shot, and must return a single self-contained HTML file.

Scores are Julian's own 0 to 10 numbers from his head-to-head guides, and medals go to the top scores per task.

Nothing on the bench grades keyword choice, factual accuracy or whether a brief holds first-hand detail.

So read every number as page-building evidence, not as proof that a model understands SEO.

Leaderboard averages read on 11 November

These averages come from the GoldieBench models page.

Averages blend games, simulations and pages, so the task-level scores below matter more for SEO work.

Two of the top names are ensembles: Fusion is described on the page as a multi-model panel, so its average reflects a system rather than one model.

For a single model you can call directly, Claude Opus 5 and GPT-5.6 Sol lead this selection, with Qwen 3.8 and Grok close behind.

Gemini 3.6 Flash sits lower on average but scores well on the page tasks, which is why the average alone is a poor guide.

Service pages: what the Landing task shows

The Landing task page lists 26 models tested, 21 scored and an average of 8.11 out of 10.

Four models share gold at 9.0: Fugu Ultra, Fusion, GLM-5.2 and Grok.

GPT-5.6 Sol scores 8.7 and Gemini 3.6 Flash scores 8.6.

The Gemini note praises the build but flags "somewhat filler AI-jargon copy".

That one comment is the whole SEO lesson: a strong layout can still carry commodity words.

Google's AI features page says there are no special optimizations needed for AI Overviews or AI Mode beyond core SEO, so the words on the page do the heavy lifting.

Treat a 9.0 build as a frame, then fill it with your process, your photos and the questions your customers ask.

Promo explainers: what the Aipbpromo task shows

The Aipbpromo task page asks for a motion-graphics video advert in one HTML file, with animated stats and an end call to action.

It shows 26 models tested, 20 scored, an average of 7.48 and Qwen 3.8 as the winner.

GPT-5.6 Sol takes silver at 8.3, Gemini 3.6 Flash scores 7.8, and Claude Fable 5 and Fusion each score 7.5.

For SEO, this is the closest proxy for an animated explainer you might embed on a results or case-study page.

Google's generative AI optimization guide says supporting text with good images and video can open more ways to appear in AI features.

How to use the scores without fooling yourself

• Use a top Landing scorer for the first draft of a service page, then rewrite the copy from your own notes.
• Use a strong Aipbpromo scorer for an explainer asset, and check every number shown in it is real.
• Never let a model invent reviews, ratings or awards for structured data or page copy.
• Keep AI-assisted pages within Google's spam policies, which treat mass-produced pages made to game rankings as scaled content abuse.

The five jobs no benchmark covers

• Crawler access. OpenAI's bot documentation says pages opted out of OAI-SearchBot won't appear in ChatGPT search answers.
• Preview controls. Google's robots meta tag page says nosnippet also keeps content out of AI Overviews and AI Mode inputs.
• Entities. Bing's Webmaster Guidelines ask for consistent names for organizations, people and places.
• Links and mentions. Paid placements need rel="sponsored", per Google's outbound link guide.
• Measurement. Search Console's generative AI reports and Bing's AI Performance dashboard show impressions and citations.

A small-business run: a wedding florist

1. Allow OAI-SearchBot and PerplexityBot in robots.txt and check the host isn't blocking them.
2. Write a brief from the owner's notes on real weddings, seasonal flowers and venue logistics.
3. Give the brief to one 9.0 Landing scorer, one shot, and keep the layout.
4. Replace every generic line with the owner's own words and photos.
5. Build a short animated explainer of a real wedding setup with a strong Aipbpromo scorer.
6. Check the Search Console generative AI report monthly for the new page.

A second run: a B2B software firm

1. Confirm the docs, integrations and comparison pages render their key text without client-side scripts.
2. Interview the product lead and two customer success staff about the questions buyers ask before signing.
3. Give a Landing scorer such as GPT-5.6 Sol (8.7) one brief per integration page, one shot each.
4. Have the product lead fact-check every feature claim line by line before anything ships.
5. Use an Aipbpromo-style animated explainer for a real onboarding walkthrough, with real numbers only.
6. Track the pages in Bing's AI Performance dashboard and Search Console's generative AI report side by side.

Notice the pattern: the model builds, the people decide what is true.

That split holds for any business size.

What no model or provider can promise

Google's hiring guidance says no one can guarantee a #1 ranking.

OpenAI's ChatGPT search article says placement is not guaranteed.

A higher bench score means a better first draft, nothing more.

The human and agency part

Goldie Agency calls itself an agency for SEO, GEO, AEO and link building, and sums up its link work as "Manual outreach, no PBNs".

Julian's own published numbers on his about page include 1,000+ campaigns and 240+ clients, and he runs the bench himself.

The agency's job is the five jobs the bench can't score, plus the judgement on which pages the models should build.

For the wider view of this work, read the GoldieBench piece on AI search optimization services.

Free extras: the 200+ AI SEO Prompts for briefs and audits, and the free link building book for the outreach side.

Also On Our Network

• Agent Operatorsthe seven-job runbook that maps each service to a stage of the AI answer
• AI Income Deskwhether a small business should pay for AI search work, and what to ask
• AI Tool Verdictfive ways to buy AI search work, scored out of 10 with reasons
• agentos.guidefive read-only Agent OS prompts that prep the AI search groundwork

Use the bench for drafts, keep people on the judgement calls, learn the workflow in the AI Profit Boardroom, and book a free SEO strategy session for the rest of your generative AI SEO services.