Ask Your AI "Are You Sure?" and Watch the SEO Advice Change

Key takeaways

• AI SEO advice is right for the average site. Your site isn't average.
• Ask any model "are you sure this is the right path for my business?" and it will usually rethink, add caveats and offer two or three alternatives.
• SEO is interpretation, not arithmetic: of Google's documentation, of experience, of algorithm updates, and of your business.
• The dangerous recommendations aren't the obviously wrong ones. They're the ones that look like best practice.
• The wrong path can cost visibility, and in the worst cases lead to penalties. Someone who knows the site has to make the call.

We use AI for almost everything at Aloha. Our technical SEO agents crawl and triage sites, and I use Claude Code every day for specs, audits and research. It makes us much faster.

It also, regularly, hands us advice that sounds exactly like best practice and would hurt the site if we shipped it. This week it happened again.

The One We Caught This Week

I was writing a technical spec for a client building a new content section on Shopify, and checking how Shopify handles duplicate URLs. On a public store, the same page loads at its real URL, /pages/our-story, and at several variants, including /pages/our-story/extra. Each copy carries a canonical tag telling Google which URL is the real one.

The AI flagged /extra as a problem and gave a clean, confident fix: point every variant's canonical at "the one real URL".

For that store's page, fine. But the section we were building puts all of its content at two-level URLs, like /pages/guides/getting-started. That second segment isn't junk. It's the page. A developer following the fix literally would have pointed every content page's canonical at its section hub, and Google could have dropped hundreds of pages from the index. Nothing would have looked broken. Every page would still load.

I pushed back because I knew how that site was built. The AI didn't, and didn't ask. When I pointed it out, it agreed immediately.

Someone without SEO experience would most likely have shipped it. The advice was clear, confident and easy to implement. That's exactly what made it dangerous.

Ask It "Are You Sure?"

Try this with any AI-generated SEO recommendation, from the latest Claude, ChatGPT or Grok model. Ask one follow-up: "Are you sure this is the right path for my business?"

In my experience, it almost always starts thinking again. It shares second thoughts, adds caveats it didn't mention the first time, and lays out two or three alternative routes. All of them sound reasonable. All of them are written with the same confidence as the first answer.

So which path should you pick?

An AI that changes its answer when you push isn't broken. It's showing you that the answer was a judgment call all along. The real question is who makes the call.

SEO Is Interpretation, Not Arithmetic

Very little of SEO is deterministic. There is rarely one provably correct answer. Most of the work is interpretation:

• Of Google's documentation: what it says, what it leaves out, and how it has changed over time.
• Of experience: what worked and what didn't, on sites like yours, over years.
• Of the algorithm: what each core update and spam policy change actually rewarded or punished.
• Of the business: which pages make money, how much risk the site can take, and what the team can realistically ship.

The model has read the documentation. It hasn't watched a site lose half its traffic after a core update, or seen which "best practice" quietly stopped working last year. And it doesn't know your site or your business unless you tell it, in detail, and even then it rarely says "this depends on something I can't see".

That's why the mistakes don't look like mistakes. They look like the answer.

The Cost of the Wrong Path

Picking the wrong path isn't neutral.

At best, it wastes engineering time on changes that do nothing. More often, it costs visibility: pages dropping out of the index, the wrong page ranking, AI answers citing a competitor instead of you. These losses are quiet. Nothing errors, and by the time the traffic report shows it, the change is weeks old.

At worst, it walks a site into a manual action or an algorithmic demotion. Scaled AI content, hidden text, doorway pages and link schemes can all be described, convincingly, as growth tactics. An AI asked to grow your traffic can make any of them sound reasonable.

Pro tip

Before you ship an AI recommendation, ask it two things: "what would this change do to my five most important pages?" and "how do you know?". The answers tell you quickly whether it understands your site or is reciting a checklist.

Before You Ship AI SEO Advice

• Test it on real pages: Apply the recommendation to your actual URLs, including the ones that make money. Does it still make sense?
• Ask what it would break: Ask the AI, then ask someone who knows the site.
• Check the evidence: Where did the finding come from, and can you confirm it a second way, ideally in Search Console?
• Be suspicious of universal advice: If a recommendation would apply to every site on the internet, it probably isn't about yours.
• Ship small first: Change one page or one template, then check the results before rolling it out.

AI Lays Out the Paths. Someone Has to Choose.

None of this is a reason to stop using AI. We use it in every audit, and it finds things faster than we ever could by hand. But it is a reason never to ship its SEO advice unreviewed.

AI can lay out the paths. Choosing the one that fits your site and your business, and won't cost you visibility or worse, still takes someone who knows the site and has seen how Google responds.