OpenTube

AI Websites Have A BIG Problem

Caleb Ulku

11,266 views

Caleb Ulku shares his thoughts on Edward Sturm's take regarding Claude sites, AI content, and scaled websites. Reaction to his take on wordy pages, search intent, and how Google treats AI writing in 2026.

Claude and ChatGPT will fill a page if you let them. This video sits with Edward’s examples — local service pages, comparison posts, above-the-fold copy — and looks at what searchers actually land on when the content was generated at scale. Eyebrows, repeated hero blocks, buried H1s, listicles, comparison tables, em dashes, images, video, bottom-of-funnel keywords, and word count all come up. So do behavioral signals, rank-and-tank traffic shapes, and the difference between a writing problem and a research problem.

Not a takedown. Edward and I have overlapped on LLM writing for local businesses before. The question is what to change before the model writes, and whether making the same generic page shorter is a strategy or just shorter slop.

Topics covered: Claude-generated websites, AI SEO in late 2026, search intent mismatches, scaled content risk, Google engagement signals, service pages vs listicles, and how much original business information has to exist before an LLM should touch the draft.

Edward's original video: Claude Sites Have a Serious SEO Problem: T...

Chapters
0:00 Opening
0:32 Cloud sites and extra words
1:23 Writing vs strategy
2:07 Traffic that spikes then drops
2:22 First site walkthrough
3:22 Above the fold
4:47 Intent on the page
5:58 Second site
6:23 AI writing tells
7:16 Lists vs one page
9:02 The useful part of the page
10:01 Scaling AI content
10:54 Images and video
12:29 Short pages and money keywords
13:05 Word count
14:09 Before you write
17:06 Closing

Keywords: edward sturm, claude ai, claude sites, chatgpt, ai content, ai seo, ai slop, seo 2026, local seo, search intent, google seo, website copy, landing pages, word count, em dashes, listicles, comparison pages, bottom of funnel, ranking factors, behavioral signals, llm writing, content strategy, scaled content, organic traffic, caleb ulku

17 Comments

@revelationsix 4 days ago

SUMMARY: You need to know what good content is, before you start using AI to generate it.

And yeah, that sounds right.

@webgeneral 3 days ago

Eyebrows are a fantastic SEO tool. I often use them to introduce a section for the robot. For example, if I want to introduce the 'About' section on the homepage, I use an eyebrow and write 'About Us.' This eyebrow is actually a disguised H2 heading. The headline below it is a regular paragraph formatted as an H2, which aims for the emotional reaction of the web user. In this way, both the robot and the web user can be satisfied.

@Contentisdata 3 days ago

At 12:54, Don't focus on Length,
Focus on STRENGTH!

@baconbigbeats 2 days ago

Here’s what Claude says about this video: Overstated or unproven

• Walls of text as a "negative sign toward indexing": Speculative. No evidence Google demotes text-heavy layouts directly. Any effect would come indirectly, through poor engagement.
• Em dashes: A human-perception tell, not a Google signal. Google says it doesn't penalize AI content as such, only unhelpful or scaled-abuse content.
• "ChatGPT focuses on listicles less now": Anecdotal. I haven't seen data confirming it.
• Duplicate hero from WP themes: Plausible and common with featured-image auto-heroes, but it's a guess about that specific site.

The main thesis is sound and matches Google's helpful content guidance: differentiated, business-specific information matters more than length, and research before prompting is what prevents slop.

@uttula 2 days ago

7:00 not using em-dashes is also ”wrong”. Know when to use them, and when not to. The plain-, en- and em-dash serve a different purpose and should be used accordingly. Of course, these days many people are not even taught what the difference is, so … it is easy to not catch it … specifically if you generate AI-slop to try and drive SEOd ranking (which IMHO should not be done and search engines should actually penalize sites for that … if they could determine it isn’t done by your competitors to drive your ranking down)

@jayyseo 4 days ago

Edward did a video with you without even knowing!

@simonkensington-fellows6142 3 days ago

Interesting that ‘eyebrows’ caught Edwards attention - Darren Shaw recently posted how he loves the ‘eyebrows’ as he uses them for the h1. Just another example of how the SEO community disagrees within itself. Which is natural and debate encouraged but problem evolves where SEO ‘clients’ only seem to catch one side of the debate - then tell you you’re not doing it right!

The broader topic of LLMs creating content is to be very careful about the brief you give, the more research you do, the more detailed the brief, the better the guardrails and QA you have the better quality you get. Same as always - garbage in - garbage out.

@bencroacademy 3 days ago

Gunna tell my agent.. I need no eyebrows at this point 🤣

@ryanmccain1158 4 days ago

watched it. good stuff

@JamalRamsey-g1z 4 days ago

Hes awesome and so are you.

@BobbyDenniegetlost 3 days ago

If you don’t prompt properly, provide enough context, set clear guardrails, and review the output like a good manager or orchestrator would, you’re pretty much guaranteed to end up in the ditch.
😂

@GaiusTacitus 11 hours ago

An AI should not be writing the text. Why not? Because material created by generative AI is ineligible for copyright, leaving the material in the public domain. Any competitor may copy and reuse the content, and there is no legal recourse available. If a contract with a customer requires delivery of all rights, then using generative AI is a material breach that leaves the consultant liable for damages.

@irinbon6207 3 days ago

look, ai is making everyone very lazy. Nobody wants to do research or move their hand, or click a mouse any more. Everyone wants a quick fix, quick task, quick solution that does not require thinking. We are sort of offloading creativity to AI, trusting that it knows better, and in the end, I think we are walking into a trap of complacency and stagnation.

@onepercentile 3 days ago

Em-dash is AI being lazy with punctuation and a dead giveaway that a human spent more time on the prompt than the actual output. If a human couldn't be bothered to write it, then a human shouldn't be bothered to read it. Human-made content, even if AI assisted, is more engaging than a perfectly formatted em-dash circus, lacking of human magic.