What Google Wants Isn’t What AI Wants

As an SEO agency, we’re currently operating in two worlds. While Google still drives the overwhelming majority of organic traffic and revenue for businesses, clients are increasingly asking questions about ChatGPT, AI Overviews, Gemini, Perplexity, and what they need to be doing to show up there.

AI is the channel everyone wants to talk about, understand, measure, and prepare for. The natural assumption is that if you are doing well in Google, you should probably be doing well in AI too.

But while plenty of overlap exists between Google and AI search, our team has learned pretty quickly that the two do not reward the same things.

That difference leaves us trying to protect what is working today while also building into another channel that is gradually becoming more important. Here’s what we’ve observed so far.

AI’s influence is bigger than its attributable revenue

For the overwhelming majority of our clients, 95%+ of revenue still traces back to Google organic, paid search, email, social, and other established channels.

AI referral traffic and directly attributable AI revenue are generally somewhere in the 1–3% range, and for a lot of businesses it is lower than that.

There are exceptions. In B2B, one lead that first discovers a company through an LLM could turn into a very large deal. When you only close a handful of contracts a year, one AI-assisted opportunity can make the channel look significant very quickly.

Across a broader client base, though, AI is not writing the checks or paying the bills yet.

Tracking revenue is important because the hype around AI can create expectations that the actual numbers do not support. 

For example, we connected with a business that generated less than $10,000 from AI-related traffic in one year, but then set a mid-six-figure AI revenue expectation for the next year. The goal was set without any trendline that suggested it was achievable.

There is just a mainstream feeling that AI is everywhere, so the revenue should be there too.

The challenge of accurate measurement

Disjointed or incomplete measurement makes tracking revenue even harder. That dark funnel has always existed in marketing, but AI adds another layer to it.

Someone may discover your brand through ChatGPT, remember the name, search for you on Google two days later, and convert through AI Mode with no clear attribution. Someone else might go directly to the site. That sale ends up attributed to organic search, direct, unassigned, or another channel even though AI played a meaningful role in the discovery.

We know that is happening. We just cannot cleanly measure how often yet.

That makes AI a difficult channel to manage and give concrete revenue attribution to. Enough is happening that ignoring it would be a mistake, but there is not enough clean revenue attribution to treat it like a mature performance channel either.

What Google wants isn't always what AI wants

A lot of the fundamentals carry across both Google and LLMs:

• Clear, consistent brand messaging matters.
• A strong About page matters.
• Your company should describe itself consistently across its website, LinkedIn, review platforms, industry publications, and other places where the brand appears.
• Blog content should be written for the audience you actually want to reach, answer useful questions, demonstrate expertise, and connect naturally back to the products or services you sell.
• E-E-A-T still matters.
• Digital PR and third-party validation matter even more when an AI system is deciding how much trust to place in what your company says about itself.

Those are all things I would recommend whether ChatGPT existed or not.

Where AI optimization starts to diverge from SEO

The strategy gets more interesting when we look at the work we are starting to do to influence how AI systems understand, retrieve, and surface a brand.

For example, we have been slowly rolling out an “AI information page” across certain clients. (Here’s ours.)

These pages explicitly state that they are intended for AI and LLM systems and provide a very clear base of information about the company: what the business is, what it sells, who it serves, how its products or services fit into the market, and other information a model needs in order to describe the brand accurately.

We are not building them as Google landing pages, and we are not trying to push users toward them through normal site navigation.

What has caught our attention is how quickly we have sometimes seen changes after publishing them. In several cases, within a few days we have seen brand descriptions become more accurate and more closely aligned with the positioning we provided. We have also seen the page itself start appearing in citations and as a source models use while researching the brand.

The results vary, but the page is one of the clearest examples we have found where something useful for AI does not really have a traditional Google SEO objective behind it.

How short-term AI tactics can create long-term risk

There is also another side to GEO that the industry needs to be careful with: tactics that can influence AI visibility today without necessarily creating useful or durable content.

Listicles

For example, self-serving listicles are working surprisingly well in AI right now. A company publishes “The 10 Best X Companies,” puts itself at the top, and that page begins influencing how an LLM understands the category.

We have watched some of these pages lose ground in Google while continuing to show up heavily in AI answers.

That tells me they are exploiting something that works in the current environment. It does not convince me that it is a good long-term strategy, especially when the content provides very little value beyond manipulating how the brand is positioned.

FAQs

FAQs are beginning to head in a similar direction as listicles. FAQs can be genuinely useful, but they are being added to everything because people have heard that LLMs like question-and-answer content. Eventually, that creates a lot of low-value content.

Misinformation

We have also seen competitors intentionally mis-position our clients in their content and then watched an LLM repeat that positioning back as fact. In some cases, it has even made its way into Google AI Overviews.

It can be tempting to fight that by doing the same thing back, but I would rather spend that effort creating better sources and stronger signals around what the company actually is. These systems will get better at identifying manipulation, and Google already has years of experience punishing the same behavior in traditional search.

Third-party authority matters more for AI visibility

One of the bigger changes for SEO teams is the growing influence of information outside the brand's own site.

Digital PR has become a much bigger part of our strategy because we want clients appearing in the third-party sources AI systems already trust.

For most businesses, that means investing in industry publications, associations, trade media, expert commentary, earned media, or wherever people in that market actually go to make decisions.

You cannot dictate every source a model chooses to trust. There are also going to be sources you cannot control. LLMs sometimes latch onto a random forum comment or an old discussion and give it far more weight than it deserves.

What you can do is create enough credible, consistent information around the brand that the model has better material to work with.

Should SEO and AI be one budget?

We currently handle SEO and GEO within the same investment for most clients.

As the work grows, there are enough AI-specific activities that I can see it becoming its own budget line. I think of it similarly to paid media: Google Ads, Microsoft Ads, and Meta may all sit underneath paid media, but they still have different budgets, tactics, expectations, and returns.

I still believe the SEO team should own AI visibility. The overlap in technical work, content, entities, digital PR, search behavior, and measurement is too strong to move it into a completely separate silo.

However, our clients know that we avoid applying mature-channel ROAS expectations to AI traffic while the attribution, revenue, and signals are still being determined.

The attribution is messy, synthetic prompt tracking has limitations, direct revenue is still relatively low for most companies, and the industry is still figuring out what some of the signals actually mean.

We can still measure progress, but the measurement needs to match the maturity of the channel in order to be meaningful.

How we measure GEO today

A big part of our work with GEO is still education. The landscape changes quickly, and that means the conversation with clients has to keep changing with it. This is our current approach to measuring GEO:

1. We track mentions and citations separately

A brand can be mentioned without its website being cited, and a citation does not mean someone clicked it. Both tell us something useful, so we do not collapse them into one metric.

2. We track share of voice across groups of prompts that matter to each client 

We run those prompts multiple times and average the results because a single response from one model does not tell you very much.

3. We watch AI referral traffic through GA4 (knowing it only captures part of what is happening) 

We look at direct and unassigned traffic for meaningful changes too, but we are careful not to give AI credit for every increase we cannot otherwise explain.

4. We watch bot activity through Microsoft Clarity 

This is a more recent experiment. I am interested in whether we eventually get to a point where bot activity becomes useful as a leading indicator of human discovery, but we do not have a reliable relationship between those two things today.

5. We compare results from multiple AI tracking tools 

Our tool list includes Semrush's AI tracking, AirOps, Knowatoa, BWT citation reporting, Search Console's limited AI data, and others depending on the client.

I would be cautious of anyone making strong revenue or timeline promises around GEO today. There just is not enough history behind the channel to support that kind of certainty.

The next problem is what happens after the citation

One of our more useful recent discoveries came from a client whose help documentation was being cited by AI systems in response to prompts from prospective buyers, not customers troubleshooting the software.

From an AI visibility perspective, it looked good. The company had earned the citation, and the LLM was sending someone to its website.

Then we followed the journey.

The prospect landed in a help documentation portal with no obvious route back to the marketing site and no way to request a demo, leaving potential customers stranded.

Once we discovered the pattern, we added a path to take that next step directly from the documentation site.

It is a small example, but it shows how AI is changing the way we think about content's role in the buyer journey. We are spending a lot of time today asking whether AI mentions a brand and whether it cites the website, but then we must look more closely at which page it chose, why it chose that page, and whether the person or agent arriving there can actually do what they came to do.

That becomes even more important as AI agents begin taking actions on behalf of users.

A prospect may eventually ask an AI to find three software solutions for a problem and then tell it to book demos with the ones it recommends. The brand will then need to be visible in the answer while also supporting the AI-driven actions that follow.

We are starting to think about that with clients now, but selectively.

There are standards and protocols still taking shape, and we won’t be chasing every announcement that gets attention on LinkedIn. When we see something with real potential, we want to be early. We just do not want being early to become an excuse for being distracted.

Our goal: prepare for what’s next while steadily investing in what works now

That is the balance between the two worlds that SEO teams are going to be managing for a while.

Google is still the stallion carrying the business for most brands. AI is the younger horse that is getting faster, getting stronger, and clearly needs attention.

You do not starve the stallion to feed the pony.

Keep investing in the channels producing meaningful traffic, leads, and revenue today. At the same time, put enough effort into AI visibility, citations, brand understanding, off-site authority, measurement, and eventually agentic readiness so that you are not starting from zero when those behaviors become more mainstream.

The proportion will be different for every company, and it will keep changing.

Our job is to keep watching where the opportunity is becoming real and push on both accordingly.