Lose Google and you lose AI search

AI search has created a new wave of tactics, tools, dashboards, and very confident advice.

Some of it is useful. Some of it is clearly going to be short-lived. And some of it looks a lot like the same SEO spam we have seen before, just wearing a new AI hat.

That is why I wanted to speak to Lily Ray.

Lily is VP, SEO & AI Search at Amsive and Founder of Algorythmic. She is one of the people I most want to hear from when the industry starts getting too excited about a new tactic, especially when that tactic comes with a dramatic case study and a chart going up and to the right. She has spent years studying Google updates, E-E-A-T, content quality, technical SEO, ecommerce SEO, news SEO, Discover, and now AI search. She is also very good at asking the uncomfortable question: did this actually work, or did it just work for a few months?

That was the shape of this conversation. Not "is AI search real?" It is. Not "does SEO still matter?" It does. The better question is: what is genuinely changing, what is being overhyped, and what should teams actually change when leadership asks whether they are showing up in AI?

˙✧˖ AI-written summary

Below is an AI-assisted summary of the webinar conversation. This is not a word-for-word transcript but is included to help you find the key parts of the conversation.

AI search is still connected to SEO

Lily started with a point that should calm some of the panic around AI search: Google has been using AI in search for a long time.

The industry has new language now: GEO, AI search, LLM optimization, AI visibility, AI Overviews, AI Mode. But Google Search has included machine learning and AI systems for years. The newer generative layer is not disconnected from the older search layer.

Lily described AI Overviews, AI Mode, Gemini, and related experiences as sitting on top of Google's search systems, indexes, ranking systems, and data. That means a lot of the work SEOs have been doing still matters.

Crawlability still matters.

Content quality still matters.

Authority and trust still matter.

Freshness still matters.

Search visibility still matters.

This is why the idea that AI search is a completely separate channel can be misleading. AI search changes the interface and the user journey, but it often still depends on the same underlying web, search indexes, and quality signals.

SearchPilot has made a similar argument in LLMs do not rank anything. So what are you optimizing for?: LLMs are not ranking pages in the old SEO sense, but when they use live retrieval, traditional search visibility and source quality still matter.

LLMs use search more than people think

One useful distinction in the conversation was between training data and live retrieval.

LLMs have training data. That is part of how they understand language, concepts, entities, and associations. But training data is not enough when the answer needs to be current.

For ecommerce, this is obvious.

A model cannot know today's price, stock status, promotion, delivery window, or latest review from old training data. It needs to retrieve that information from somewhere.

That is where search comes back in.

Will pointed out that this is one reason search is not going away. AI systems need up-to-date information. Whether the interface looks like a traditional results page, an AI Overview, a chat answer, or an AI shopping assistant, the system still needs access to current facts.

Lily connected this to old Google ideas like QDF, "query deserves freshness", and the push Google made years ago to crawl, index, and process the web faster. The details may be different now, but the underlying need is familiar: fresh information matters when the world changes.

That is especially true for ecommerce teams, where pricing, stock, reviews, product drops, and availability can change quickly. It also connects to SearchPilot's Merchant Center Testing, because product feeds, structured data, and product pages all become part of the information layer that search and AI systems use to understand what is available.

Lose Google and you may lose AI search too

One of the strongest ideas from the session was Lily's warning that the fastest way to lose AI visibility may be to lose Google visibility first.

The logic is straightforward.

If AI systems use search results to retrieve fresh information, then the sites that lose visibility in Google may also become less visible in the AI systems that depend on Google, Bing, or other search infrastructure.

That does not mean every AI answer is simply a repackaged search result. Training data, brand associations, mentions, and other sources still matter. But in many practical cases, AI systems use search to answer current questions. If a site becomes less visible in that retrieval layer, it can lose citations, recommendations, and inclusion in AI answers.

This matters because a lot of current GEO tactics are designed to influence AI answers quickly, without enough thought about how those same tactics will look to Google later.

Lily's warning was that this can create a cycle:

A tactic influences LLM responses for a while.

The tactic becomes public.

More people copy it.

A clear footprint appears.

Google or another search engine catches up.

The site loses organic visibility.

AI citations and recommendations decline because retrieval visibility declines.

That is why "lose Google and you lose AI search" is not just a provocative title. For many brands, especially those relying on fresh retrieval, it is a real risk.

Some GEO tactics look like old spam

Lily has been watching a new set of companies and consultants enter the search industry through GEO and AI search.

Some of them do not have much SEO history. That matters because they are sometimes repeating patterns that experienced SEOs have seen before.

Mass content.

Doorway-style pages.

Scaled pages targeting slight variations of the same intent.

AI-generated pages built because they can be built quickly.

Public case studies showing short-term spikes.

All of that may look new because the surface is new. But the pattern is old.

Lily compared some current GEO tactics to the kinds of content strategies that were hit by older Google systems, including Panda, the helpful content system, and other quality updates. The difference now is that AI makes production dramatically easier. A team can generate thousands of pages in minutes.

That does not make the strategy durable.

It may work for a short time, just as spammy SEO tactics often worked before Google introduced countermeasures. But working briefly is not the same as being a good strategy.

This is where the Omio example is useful. SearchPilot's GEO A/B testing customer story shows that SEO and GEO are not always the same. One Omio test increased LLM traffic by +18%, while another GEO-positive change would likely have hurt Google organic sessions by -6.5%, so it was not rolled out.

That is the point: teams need to understand the trade-off, not assume that every AI visibility tactic is safe for organic search.

Case studies need a longer shelf life

Will and Lily also talked about how public some AI search case studies have become.

That is unusual in SEO.

Experienced SEOs know that if a tactic is genuinely aggressive and works very well, shouting about it in public is not always wise. Lily made the point that Google, Microsoft, and other companies are watching. Even when they are not watching a specific case study, public promotion makes it easier to identify the footprint once enough sites copy the tactic.

Lily has been tracking this more systematically.

She described building dashboards to monitor hundreds of companies that have been named in public AI search or GEO case studies. What she has often seen is that the case study is published near the peak of the traffic or citation curve. Then, in the following months, some of those sites see a sharp decline.

The case study remains live, but the long-term result looks very different.

That should make teams cautious.

A case study can be useful. It can suggest a hypothesis. It can show that something had an impact in one context. But it should not become a playbook without testing.

The better question is not "did this work for someone on LinkedIn?"

The better question is "would this work for our site, with our constraints, without hurting the channels we still rely on?"

SearchPilot's GEO A/B Testing is built around that mindset: test AI visibility changes while measuring the effect on traditional organic search, AI referral traffic, and the overall business outcome.

AI-assisted content is not automatically bad

The conversation was not anti-AI content.

Lily was clear that AI-assisted content can be useful, especially when it is summarizing, structuring, or synthesizing real information.

There are ecommerce use cases where AI-generated or AI-assisted content may be perfectly fine. A product description, a summary of specifications, a review summary, or a rewritten explanation of existing data can all be useful if the source material is accurate and the output helps customers.

The problem is scale without value.

There is a big difference between using AI to turn proprietary product data, reviews, research, or expert insight into something more useful, and asking AI to create thousands of generic pages because it can.

The first can help.

The second starts to look like the same low-value content Google has fought for years.

Will raised a related point: AI may eventually become much better at producing writing that people actually enjoy. It is already useful for summarizing long pieces, comparing multiple sources, and producing drafts. But Lily pushed back on the idea that AI content is already equivalent to human writing in voice, wit, experience, or originality.

Her point was simple: reading something written by a person still feels different.

That matters for brands. The more the web fills with AI sameness, the more genuine human perspective, expertise, humour, testing, experience, and originality may stand out.

SearchPilot's AI content testing roadmap makes the same practical point: AI content should be tested, not assumed. Some uses will help. Some will do nothing. Some may hurt.

AI search spam is already a problem

Will asked Lily what she is seeing from the model and platform side: hallucinations, misinformation, spam, prompt injection, and adversarial information retrieval.

Lily's view was that the AI companies have a lot of work to do.

She has been surprised by how easy it can be to get AI-generated content, including poor-quality AI-generated content, cited by AI systems or used to influence answers.

That creates a loop.

Bad AI-generated content gets published.

Another AI system cites or summarizes it.

More AI-generated articles repeat the same claim.

The idea starts to look more legitimate because it appears in more places.

Lily gave GEO advice itself as an example. A lot of AI search advice online is not rooted in strong evidence. But if enough people write about it, and AI systems ingest or retrieve it, that advice can become institutionalized even if it is not true.

Will compared this to an old reputation management problem with Wikipedia. Someone adds a questionable claim to Wikipedia. A journalist repeats it in an article. Then Wikipedia cites the article as evidence. The loop closes, and the claim becomes harder to remove.

The AI version of that loop may be faster and larger.

That is why platform-level spam detection, source quality, and adversarial thinking matter. It is also why SEO experience matters. Search engines have spent decades fighting people trying to manipulate retrieval systems. AI companies are now speed-running some of those same lessons.

Prompt tracking is directional, not definitive

Prompt tracking is one of the fastest-growing areas around AI search.

Many brands are now tracking whether they appear in ChatGPT, Perplexity, Gemini, AI Overviews, AI Mode, and other AI surfaces for a set of synthetic prompts. That can be useful. It gives teams a view of whether they are present in important AI answers and how they are being described.

But Lily was careful about the limits.

Prompt volume is directional at best. The data is sampled. The tools are early. The prompts are synthetic. AI answers vary. Personalization is increasing. Follow-up questions change the context. User history can change the response.

In other words, a prompt tracker is not a perfect map of demand.

Will described the extreme version of this as "search volume one." If people ask long, specific prompts, and those prompts are shaped by personal history and context, then many AI searches may be effectively unique.

That makes measurement hard.

Lily made an interesting counterpoint: this may make traditional keyword search volume more useful than people think. Search volume for core keywords may still be one of the better sources of truth because it captures the underlying intent at scale, even if the AI prompt itself becomes personalized and fragmented.

The practical takeaway is not to ignore prompt tracking. Lily uses it. Amsive uses it. Many SearchPilot customers use it.

The takeaway is to treat it as directional evidence, not certainty.

For teams trying to understand what AI referrals look like after the click, SearchPilot's How does AI traffic show up in analytics? is a useful companion piece.

Do not track 5,000 prompts just because a tool lets you

Will and Lily also discussed the economics of prompt tracking.

For an ecommerce site with hundreds of thousands of SKUs, it is tempting to think the answer is simply to track more prompts. More products, more modifiers, more use cases, more comparison phrases, more categories.

That can quickly become expensive and noisy.

Lily referenced research suggesting that many differently worded prompts asking essentially the same thing may return similar answers. That points toward a more practical approach: cluster prompts around core customer questions rather than trying to track every possible wording.

This is similar to keyword clustering.

The exact wording varies, but the underlying intent may be the same.

For large ecommerce teams, that means the prompt tracking strategy should start with commercially meaningful clusters:

• brand comparisons
• category-level discovery
• product suitability questions
• "best for" use cases
• buying guides
• high-margin product groups
• high-volume categories
• trust and reputation queries
• post-purchase or service questions

The goal is not to create the biggest possible dashboard.

The goal is to learn what to change.

Agentic shopping is early, but real

Will asked Lily whether agentic shopping is hype or whether something real is happening.

Her answer was balanced: it is real, but it is early.

Amsive's technical SEO team is already digging into emerging standards and concepts around agentic commerce and agentic search. But Lily also said it will take time before average consumers are comfortable letting an agent shop for them.

For mundane purchases, the shift may happen sooner. Reordering toothpaste or a household staple feels different from letting an agent choose an expensive outfit, a sofa, or a holiday.

Google's advantage here may be personalization. Lily pointed out that Google has access to years of purchase history for users whose receipts and order confirmations live in Gmail. In AI Mode, that can make product recommendations feel much more personal: find an outfit for a conference, find shoes that match past preferences, find something similar to what the user has bought before.

Will's view was that the journey may split into two parts.

The research and comparison stage becomes more agentic.

The checkout stage becomes easier, almost like "fancy Apple Pay."

The user may still want to approve the choice, see the product, compare a few options, or visit the retailer. But once they decide, nobody particularly enjoys filling in forms and payment details.

That makes the near-term ecommerce question less about fully autonomous shopping and more about how retailers appear during AI-assisted research and comparison.

SearchPilot's recent article on Agentic Commerce SEO goes deeper on what ecommerce teams should test now across PDPs, PLPs, product feeds, structured data, content blocks, and Merchant Center.

Links still matter

Will asked Lily whether AI search experiences will move back toward more links.

He thinks they will.

Perplexity is already link-heavy. Google has faced pressure from publishers, recipe sites, and media companies to send more traffic and create a healthier web ecosystem. Lily agreed that AI search results are likely to become more interactive, with more links, images, sources, and options to click.

She mentioned Google's Web Guide direction as a possible shape of things to come: a blend of traditional search and AI search.

But she also added an important caveat.

More links do not automatically mean more clicks.

User behaviour is changing. Some people prefer to get the answer in the AI interface. Some journeys may happen through voice. Some users may never need to click if the answer is simple enough.

That is the hard part for SEO teams. The presence of links does not guarantee the old traffic model returns.

This connects back to SearchPilot's piece on When is a click not a click?. A click in the AI era may come later in the journey, after more research has already happened. It may be fewer in number but higher in intent.

For ecommerce, that makes product pages and product feeds more important, not less. If the click happens later, the page has to do more work when it arrives.

Ecommerce teams still need the fundamentals

The session repeatedly came back to the same point: ecommerce teams should not abandon SEO fundamentals.

AI discovery makes the fundamentals more important because AI systems need something reliable to retrieve, interpret, and summarize.

That includes:

• crawlable product pages
• clean internal linking
• structured data
• accurate product feeds
• useful category and product content
• freshness signals
• reviews and trust signals
• availability and delivery information
• strong brand and product authority
• fast, usable pages

For ecommerce teams, product detail matters because AI systems may need to compare products on specifics: size, material, use case, price, fit, stock status, warranty, returns, sustainability, reviews, and delivery speed.

This is why SearchPilot's Merchant Center Testing is relevant to AI search as well as Google product listings. Product pages, feeds, schema, and metadata all contribute to whether products can be found, compared, and clicked.

What teams should focus on now

The most useful shift is from "Are we showing up in AI?" to "What should we actually change, and how do we know it is working?"

That question avoids two traps.

The first trap is chasing every new GEO claim. There is too much noise, too many dashboards, and too many confident recommendations based on thin evidence.

The second trap is pretending nothing has changed. AI search is changing discovery, summarization, comparison, and the meaning of a click.

The practical focus should be:

First, protect Google visibility. It still matters directly, and it may feed AI visibility too.

Second, improve crawlability, freshness, and source quality. AI systems need current, reliable information.

Third, be careful with scaled AI content. Use AI to improve useful information, not to flood the web with low-value pages.

Fourth, use prompt tracking as directional evidence. Do not treat synthetic prompts as a perfect demand map.

Fifth, measure downstream traffic and business impact. Visibility is interesting, but leadership needs to know what changed.

Sixth, test changes before rolling them out at scale.

That last point is where SearchPilot's GEO A/B Testing becomes important. The method is designed to measure AI search and organic search together, so teams can see whether a change helps one surface, hurts another, or creates a net positive outcome.

Put Search in Control Mode with SearchPilot

This conversation with Lily was a useful reminder that AI search is not a reason to forget everything SEO has learned.

Search has always been adversarial. Google has always had to deal with spam. Marketers have always looked for shortcuts. Case studies have always looked best before the update catches up. The difference now is that AI makes both the opportunity and the risk move faster.

For enterprise ecommerce teams, the answer is not to guess harder or chase every new AI search tactic.

The answer is to test.

SearchPilot helps enterprise teams make SEO and GEO testable. We run controlled experiments across product pages, category pages, templates, navigation, internal linking, content, structured data, Merchant Center surfaces, and AI-influenced journeys, then help teams understand what moved and what the result means.

The Omio story shows why this matters. SearchPilot found a GEO change that increased LLM traffic by +18%, and another that performed positively for LLM-driven traffic but would likely have hurt Google organic sessions by -6.5%. Without testing, those trade-offs would have been guesswork.

For more examples of controlled SEO testing in practice, see the SearchPilot customer stories, including work with ecommerce, marketplace, travel, and enterprise brands.

Search is your biggest channel and least understood. AI search does not make that less true. It makes control more important.