You’re probably being asked to do something about AI, AEO, or GEO. The question may be framed as AI search, AI, or AI visibility. The wording changes. The pressure behind it's usually the same.

You need to know if buyers can still find and understand your business when AI tools like ChatGPT and Google Gemini are part of their research process.

That pressure is reasonable.

It's also easy to misread.

The CMO Survey says pressure from CEOs, boards, and CFOs is still part of life for most marketing leaders. It also says 70.6% of marketers respond to that pressure by focusing on short-term impact over long-run gains. (View PDF Report) That matters because AI discovery is landing in a setting where visible movement can feel safer than slower, better work.

AI has also moved past the experiment stage for many marketing teams. Spencer Stuart reports that more than three quarters of surveyed marketing leaders have started piloting AI projects. Nearly half have started scaling proven use cases.

So the concern is valid. You really should be asking how your site shows up when people use AI to research.

But urgency is not the same as direction.

The right response is to ask what the work will improve on the site, and if that change helps buyers understand your offer or verify claims.

What should AEO and GEO work actually improve?

AEO and GEO work should make your website easier for buyers and AI systems to understand. The work may improve a page, a template, a module, a content relationship, or a site-wide setting. It should solve a clear problem and make the site easier to access, understand, verify, or use. If it only adds FAQs, schema, reports, or new pages without doing that, it is probably activity, not progress.

AI discovery is bigger than a search engine

Google still matters. But it shouldn't be at the centre of every AI conversation. It's a big player, but just one of them.

In Cloudflare's May 2024 to May 2025 comparison, Googlebot was still the largest crawler in the group it studied. At the same time, ChatGPT-User requests rose 2,825 percent! and GPTBot requests rose 305 percent.

OpenAI documents separate crawlers and user agents for different product uses. Perplexity does the same. Its documentation separates PerplexityBot from Perplexity-User.

A crawler request doesn't prove buyer intent.

It shows that your website is being read in more ways than a traditional search results page. Your page now has to work for the buyer who reads it and for the systems that may summarize it before the buyer visits.

That's the shift.

Your website now needs to work for search, AI systems, and the buyers who use both.

AEO and GEO are useful only if they improve the page

• AEO means answer engine optimization. In plain terms, it should make your content easier to answer from.
• GEO means generative engine optimization. It should make your content easier for AI systems to retrieve, explain, and cite.
• AI visibility is the outcome people want. It means your business appears when AI systems help someone research a topic or compare options.

Those terms are useful when they help us describe how discovery is changing.

They become a problem when they are used to rename shortcuts.

Google’s own guidance is useful here, even though Google is only part of the picture. For Google Search, the company says AEO and GEO are still tied to SEO. It also says there is no special Schema.org markup required for generative AI search. It says you do not need to rewrite content in a special way just for AI systems.

That doesn't make AEO or GEO fake.

It means the work cannot stop at markup or new answer sections.

AEO and GEO are not separate layers to add. They are a test of whether the page does its job.

Weak advice often looks responsible

Experienced teams are not immune to weak advice because they are uninformed.

The risk is different.

You may already know enough to be skeptical. You may have seen shallow SEO packages before. You may know that publishing more doesn't automatically mean improving more.

But AI discovery is still a moving category. The language is new. The reports look technical. Leadership wants to see progress. Vendors know that.

That's how weak work gets approved.

• A new FAQ block looks like action.
• A schema update looks technical.
• An AI visibility report looks measurable.
• A batch of new pages looks like progress.

Any of those things can be useful.

None of them are useful by default.

Technical-looking activity is not proof that the work will improve the website.

The shortcut is mistaking visible output for website improvement

An FAQ section can help when it answers a question a real buyer would ask on that page.

It fails when the questions are generic. It fails when the answers repeat what the page already says. It fails when the section exists only because someone believes AI systems want more answer-shaped content.

AI-assisted content has the same line. It can help a team explain something more clearly. It can also create low-value pages at scale.

The issue is not if AI helped create the content. The issue is whether the content helps the reader. Google defines scaled content abuse as creating many pages mainly to manipulate rankings and not help users. The policy says this applies no matter how the content is created.

Schema has a clear limit too.

Structured data can help systems understand what is on a page. It doesn't create proof. It doesn't create reviews. It doesn't create expertise. Google’s structured data guidelines say not to mark up content that's not visible to readers of the page.

If the page does not support the claim, the markup shouldn't imply it.

FAQs, schema, and AI-assisted content only help when they improve the page a buyer needs to use.

Some AI manipulation is already being documented

Most weak AI discovery work will look harmless. It may appear as new content, extra markup, or another report.

Some manipulation is more direct.

Microsoft security researchers have reported a tactic they call AI Recommendation Poisoning. In the cases described, companies embedded hidden instructions in “Summarize with AI” buttons. Those instructions tried to make an AI assistant remember or recommend a company in the future. Microsoft said it found more than 50 unique prompts from 31 companies across 14 industries.

That's not optimization.

It's an attempt to bias the system.

Most website owners will not be offered something that obvious. The more common risk is paying for work that looks active but leaves the page no stronger. Sound familiar?

Make the website easier to understand. Do not try to manipulate what an AI system says.

The issue is context, not where the provider sits

Low-quality AI discovery work can come from anywhere, not just offshore.

The issue is not geography. The issue is whether the provider understands your business well enough to improve the page, not just produce more assets.

The work needs to reflect what you sell. It needs to reflect what the buyer is trying to decide. It needs to use the proof you actually have. It also needs to fit what your team can maintain.

Without that context, the work becomes production.

More pages. More markup. More reporting.

That may create work to review. It may not create a page that explains the business better.

The provider needs enough business context to improve the page, not just produce more assets.

Useful AI discovery work starts by asking what the page needs to do

Start by identifying what needs to change and why. It may be a page that does not answer an important buyer question. It may be a template that produces inconsistent service pages. It may be a module that needs to handle proof or FAQs consistently. It may be crawler access, internal linking, or schema that should be managed site-wide.

The implementation should match the problem. A one-off page edit may be right. If the same improvement applies across a page type, it should usually be handled in the template, module, global group, or site-wide configuration.

They also matter to AI systems because the page has to be parsed and interpreted before it can be useful as source material.

This is where AEO and GEO earn their place.

They give us a practical reason to fix issues that often sit too long. Thin service pages. Vague claims. Weak proof. Disconnected FAQs. Similar pages that compete with each other. Content that reflects internal language instead of buyer questions.

That's the value of doing the work properly.

AI discovery is not a reason to write around people. It's a reason to make the page easier to read, explain, and trust.

Start with the buyer decision the page needs to support, then choose the tactic.

Use one standard before you approve the work

A useful AI tactic should be to improve the website even if the AI systems didn't exist.

Before approving the work, ask what needs to change, why it matters, and where it should be implemented. Is this a page edit, a template change or module update, a global setting, or a sitewide technical task? What will it improve? What will your team need to maintain?

If the provider cannot explain the problem, the intended outcome, and the right implementation level, do not approve the work.

AI discovery is real.

The opportunity is real.

But the goal is not to make a weak website look optimized for AI.

The goal is to make your business easier to understand, trust, cite, and choose.

If the tactic does not improve a specific page, do not approve it.