99% of AI + SEO Tools & Prompts are useless. Context still reigns supreme
SEO Context still reigns supreme. Illustration by ChatGPT

99% of AI + SEO Tools & Prompts are useless. Context still reigns supreme

The false promise of “plug-and-play” AI SEO

AI is supposed to save time, boost rankings, and automate tedious SEO work. But in reality, most AI-driven SEO tools and prompts you see on LinkedIn or in ChatGPT prompt libraries fall flat. Not because the AI used is bad, but because they lack one thing: context.

I have been guilty so it is time to improve. A few common problems to show you we need to alter our thinking:

1: “Contextless” link recommendations are risky

Most AI link tools recommend anchor text and link targets based solely on semantic and topical text similarity.

The flaw: it lacks context about:

  • Current internal link structures
  • Business priorities (e.g., commercial vs informational pages)
  • Authority flow (PageRank distribution)
  • Cannibalization risks

Example: A tool might suggest linking from a blog to a product page, but if that page already has strong link equity and another needs a boost, you're diluting value and not enhancing it. How about a product out of stock?

Better approach: Use AI to create an app or script (it needs to calculate, so don't use an LLM!) simulate internal linking scenarios based on crawl data and business goals, not just surface-level NLP.

2: Content rewriting with no awareness of existing assets

Tools often rewrite content based on surface-level optimization (e.g., matching a SERP snippet or competitors content) but don’t check:

  1. Is similar content already on your site?
  2. Is this a duplicate or cannibalization risk?
  3. What’s already performing organically?

Example: An AI prompt might rewrite a “how to apply for a visa” guide without realizing there's a similar article in a subfolder resulting in splitting link signals and keyword targeting.

Fix: Connect your AI rewriting prompts with your content inventory and performance data. Create context first, recommendations second.

3: Blind optimization based on current rankings is lazy

Using AI to analyze top-10 SERP results and copy their structure is common.

The myth: If it works for them, it will work for you.

Reality: You’re not factoring in:

  • Domain authority differences
  • Content freshness
  • Search intent variation
  • Alternative formats (video, calculators, etc.)
  • User signals like Navboost

I've been very guilty of this. Also see my session from SMX Advanced in 2023: https://www.notprovided.eu/dealing-with-user-intent-in-a-time-where-google-depends-on-ai-smx-advanced-berlin-2023/ about dealing with user intent in Google's volatile SERPs. Intent has become personal in many search interfaces. We need to cater for more scenario's nowadays.

4: Static keyword tools ignore new and emerging queries

15% of all queries are new and yet most AI tools are glued to historical keyword volume and static data sources. The same with prompt libraries.

Predictive content creation > reactive content recycling.

Example: Google Trends + prompt engineering can reveal emerging angles. New content ranks easily because no one has written about it. Have a good laugh (and think about the consequences) about the experiment by Lily Ray https://lilyray.nyc/which-ai-search-tools-llms-are-the-most-gullible/

This test demonstrates how quickly and easily internet-connected LLMs can potentially be influenced by newly indexed content, even if that information is not entirely true or reliable.

Future-forward idea: Combine trending data and AI prompt chains to generate content for tomorrow’s queries, not yesterday’s.

5: Great content alone doesn’t rank without authority

Most AI SEO tools act like content is king. But without:

  • Citations
  • Links
  • Brand signals

... it’s invisible.

Example: A perfect 2,000-word guide on a niche topic won’t rank if five better-known sites publish something 70% as good.

Fix: Combine AI-generated content with an off-page strategy: digital PR, link earning tactics, and citation building.

6: LLMs can’t crawl or interpret your site like traditional search engine do

Large language models like ChatGPT or Claude are not crawlers. They don't know your actual site structure, indexation status, canonicals, hreflang setup, or how Googlebot or Bingbot sees your pages.

ChatGPT (in 2024) even hallucinated links to their content partners: https://www.niemanlab.org/2024/06/chatgpt-is-hallucinating-fake-links-to-its-news-partners-biggest-investigations/ and it seems they haven't improved a lot. That 3% is in line with Ahrefs findings that show

"3.6% of AI search traffic to Ahrefs was to hallucinated links."

https://ahrefs.com/blog/ai-search-traffic-by-page-type-ahrefs/

The issue: Most prompt-based SEO tools treat websites as isolated text blobs. But real SEO is deeply structural:

  • Duplicate content from parameterized URLs?
  • Orphaned pages that aren’t linked?
  • Pages blocked in robots.txt or with noindex?
  • Incorrect canonical or hreflang tags?

Example: You ask an LLM: “How can I improve this page?” It gives text suggestions without knowing it’s already excluded from indexing or has five duplicates across subdomains.

What does improving even means? What are you trying to improve?

Bottom line: Without crawl data, log files and index status, an LLM’s advice is surface-level at best and completely wrong at worst.

Fix: LLMs must be paired with tools like Firecrawl, Screaming Frog, Sitebulb, or your own crawl exports to get any SEO-meaningful insight.

Is all AI SEO useless? No: but we’re using it wrong at the moment

The problem isn’t the AI. It’s the oversimplified use of it in prompts and tools that ignore what makes SEO hard in the first place: context, trade-offs, and competition.

We are in an early stage: give our industry time to catch up. There are tools like Similar.ai that have been doing this for years by providing context first so it is not impossible.

What we actually need

  • A data-driven approach rooted in strategy and performance
  • Integrated SEO + business logic, not siloed prompt templates
  • Custom AI workflows, trained on your domain, competitors, and goals
  • Experimental mindset, not checklist SEO

Final thought

Until LLMs can see, crawl, and understand your site like Google does and until tools account for content, links, citations and technical structure in context, most AI-powered SEO tools are just fancy autocomplete machines.

The promise is huge. But we’re not there yet.

Gus Pelogia

Sr. SEO & AI Product Manager @ Indeed | Search ‘n Stuff and SEMRush Spotlight Speaker (Oct ’25)

3mo

Love the idea to simulate internal linking. Wondering how to build something to do it (with my current skills)

Like
Reply
Asim Hussain

Expert in 🖇️ Link building, Guest Posting & onpage content writing

4mo

Love this, Jan-Willem

Like
Reply
Risto Rehemägi

Co-Founder at ContentGecko | AI SEO for WooCommerce stores

4mo

It's all about giving AI the correct context. With ContentGecko, you can map out every little detail about your domain and SEO strategy and THEN use it to help you write.

  • No alternative text description for this image
Josep M Felip

SEO Specialist | AI Cowboy | SEO Mentor | Search Awards Judge

4mo

Thanks for sharing Jan. Totally agree, just giving prompts without context does not work. You can’t expect to outrank competitor without fresh information, original angles and the right EEAT signals.

Like
Reply

To view or add a comment, sign in

Others also viewed

Explore content categories