⭐ Get the Agent OS + join 3,400+ founders inside the AI Profit Boardroom → Join AIPB

GoldieBench Blog · 6 min read

AI SEO Company Tasks: Which AI Models Draft The Pages, Tools And Assets, With Real GoldieBench Scores

Live GoldieBench scores for page drafts, on-page tools and launch assets, mapped to the deliverables an AI SEO company produces.

Illustrated benchmark leaderboard with gold score bars beside web page wireframes and a schema card on a dark navy background

Much of what an AI SEO company produces is drafting work, such as service pages, schema, briefs and interactive tools, and AI models now handle a big share of those first drafts.

This page matches those tasks to the closest GoldieBench results, using only scores read on live goldiebench.com pages on 11 Nov 2026.

For the strategy, links and judgement no model covers, book a free SEO strategy session with my team.

What the bench measures, and what it doesn't

The GoldieBench leaderboard lists 50 tasks, 33 models and 1,310 live demos, split into Games, Visuals, Sims and Pages.

Rankings come from Julian's 0–10 scores, averaged across every task a model ran.

Pages, with 3 tasks, is the category closest to SEO deliverables.

No task on the bench tests schema, content briefs or outreach research, so treat the scores as a guide to page-building skill, not to SEO judgement.

Mapping AI SEO company tasks to the bench

SEO deliverableWho does itNearest GoldieBench task
Service and landing pagesModel drafts, person adds proof and factsLanding
Calculators, quizzes and other on-page toolsModel builds, person testsWeb-OS Desktop
Promo assets for launchesModel drafts, person checks claimsAIPB Promo
Organization or LocalBusiness schemaModel drafts, person matches it to the pageNone
Content briefs and outreach listsModel researches, person decidesNone
Crawler access, vetting and reportingPeopleNone

Landing task: who drafts the best service pages

The Landing task page gives each model one shot at a modern marketing landing page.

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

ModelScore on goldiebench.com/tasks/landing
Fugu Ultra9.0/10 (winner)
Fusion9.0/10
GLM-5.29.0/10
Grok9.0/10
GPT-5.6 Sol8.7/10
Gemini 3.6 Flash8.6/10
Claude Fable 57.2/10

Any of the four 9.0 models makes a strong first pass at a service page.

The draft still needs your real services, real photos and first-hand detail before it goes live.

Web-OS task: who builds working on-page tools

Interactive tools such as cost estimators can make a page worth citing, and the Web-OS Desktop task is the working-app build in the Pages category.

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

ModelScore on goldiebench.com/tasks/webos
Fusion9.0/10 (winner)
Grok9.0/10
GPT-5.6 Sol8.4/10
Claude Fable 58.0/10
Gemini 3.6 Flash8.0/10
GLM-5.27.5/10

Fusion and Grok top both Landing and Web-OS, which makes them a sensible pair for page builds.

AIPB Promo task: who handles launch assets

The AIPB Promo task asks for a cinematic motion-graphics advert in a single HTML file.

It reports 26 models tested, 20 of 26 scored, an average of 7.48/10, and Qwen 3.8 as winner.

Scores read on that page include GPT-5.6 Sol at 8.3/10 (silver), Gemini 3.6 Flash at 7.8/10, and Claude Fable 5, Fusion and GLM-5.2 at 7.5/10 each.

GPT-5.6 Sol is the steadiest all-rounder here: 8.7 on Landing, 8.4 on Web-OS and 8.3 on AIPB Promo.

The overall leaderboard

On the leaderboard page, Fusion leads with an 8.59 average, followed by Claude Opus 5 at 8.27, Hermes MoA at 8.17 and GPT-5.6 Sol at 8.16.

The leaderboard tags GPT-5.6 Sol, GLM-5.2 and Gemini 3.6 Flash as "best at Pages".

Overall averages mix games and simulations, so for SEO page work the task-level tables above are the better guide.

Where the platform docs matter more than any model

A model can draft perfect HTML and still miss AI answers if the basics are wrong.

  • Google says pages need to be indexed and snippet-eligible to show in AI Overviews, with no extra requirements, per its AI features page.
  • OpenAI says ChatGPT search won't show sites that opt out of OAI-SearchBot, per its crawler docs.
  • Perplexity says to allow PerplexityBot to appear in results, per its crawler page.
  • Anthropic notes that disabling Claude-SearchBot can lower how often you show in Claude's search results, per its crawler article.
  • Bing says Copilot shares the same crawl and index foundation as Bing search, per its Webmaster Guidelines.

Google's generative AI guide also warns that mass-producing variant pages to catch fan-out queries can break its scaled content abuse policy, however fast a model writes them.

Gotchas when a model drafts SEO pages

  • Invented facts. A one-shot page will happily make up awards, review counts or service areas; strip anything you can't prove.
  • Text locked inside scripts. Google's AI features page says important content should be available as text, so check the rendered HTML shows the copy.
  • Schema that drifts from the page. Google asks for structured data that matches the visible text.
  • Commodity copy. Google's guide favours first-hand detail, and a model can't supply your job photos or customer stories.
  • Accidental preview blocks. A template that ships nosnippet keeps the page's text out of AI Overviews, per Google's robots meta tag docs.

A strong benchmark score buys you a faster first draft, not a finished page.

A model pairing for an AI SEO workflow

  1. Page drafts: a Landing 9.0 model such as Fusion or Grok.
  2. Interactive tools: Fusion or Grok again, both 9.0 on Web-OS.
  3. Second opinion and edits: GPT-5.6 Sol, consistent across all three Pages tasks.
  4. Schema, briefs and outreach research: any strong model, checked by a person, since no task scores them.

The human and agency part

Models draft; people choose topics, add first-hand proof, earn links and approve every live change.

Goldie Agency runs that side as an SEO, GEO, AEO and link building agency, with outreach described as "Manual outreach, no PBNs".

Going by Julian's own published figures, his channel has 425K+ YouTube subscribers and the agency lists 240+ clients, over 1,000 campaigns and a 100% Upwork Job Success Score.

No agency can guarantee a top ranking; Google's hiring guidance says so plainly.

Small-business example: an electrician's website refresh

  1. Draft five service pages with a Landing 9.0 model, then the electrician adds real job photos and certifications.
  2. Build a simple rewiring cost estimator with a Web-OS 9.0 model and test it on a phone.
  3. Draft LocalBusiness schema and check the phone and address match the contact page.
  4. Confirm robots.txt lets OAI-SearchBot, PerplexityBot and Claude-SearchBot in.
  5. Hand outreach to an agency and review its list before anything is sent.

For how firms differ on this split, see our look at what an AI SEO firm delivers.

Free extras: the 200+ AI SEO Prompts for briefs and audits, and the free link building book. Ready to talk? Book a free SEO strategy session, or learn AI SEO yourself in the AI Profit Boardroom.

Also On Our Network

Let the top-scoring models draft, keep people on facts and links, and book a free strategy session when you want a proven AI SEO company on your side.

FAQ

Which model scored highest on the GoldieBench Landing task?

Fugu Ultra is named winner on goldiebench.com/tasks/landing at 9.0/10, tied on score with Fusion, GLM-5.2 and Grok.

Does GoldieBench test schema or content briefs?

No. The Pages category has three tasks, Landing, Web-OS Desktop and AIPB Promo, so schema, briefs and outreach have no direct score.

Which model is the most consistent across Pages tasks?

GPT-5.6 Sol scored 8.7 on Landing, 8.4 on Web-OS and 8.3 on AIPB Promo on the live task pages.

Can a high-scoring model replace an AI SEO company?

No. Models draft pages, but crawler access, first-hand proof, link earning and approvals still need people.

The same stack Julian uses

Run this stack yourself.

Every demo on this bench was built inside the Agent Operating System — one prompt, one shot, single HTML file out. The Agent OS, the prompts, the templates, the weekly walkthroughs and 3,400+ founders shipping with it every day all live inside the AI Profit Boardroom.

3,400+founders
258documented wins
38countries
4coaching calls a week