Crawl, Walk, Run: How to Sequence Your AI Search Program

In short: AI search programs should progress through three stages: Crawl to establish accurate visibility, Walk to measure and attribute performance, and Run to evaluate traffic quality and business impact. This sequence gives teams a way to prioritize GEO work based on what they can reliably measure today. 

Key Takeaways

• AI search programs work best in the Crawl, Walk, Run sequence because each maturity stage depends on the one before it.
• Crawl establishes whether your brand is present and accurately represented in AI search.
• Walk focuses on growing that presence and building a defensible approach to measurement and attribution.
• Run determines whether AI-referred visitors are the right audience and whether they contribute to business outcomes.
• Large organizations may be at different stages across products, regions, or business units. The goal is to know where each part of the business actually stands.

Marketing teams are moving fast on AI search, but often without a reliable order of operations. A 2026 Semrush study found that only 22% of marketers have fully integrated AI search and SEO into their operations, with strategy often moving faster than the operating model behind it. 

Teams are often tracking prompts, publishing content, debating attribution, and trying to prove business impact all at once. That makes it hard to tell what’s actually working

If you’re not getting traffic yet, attribution won’t tell you much. And if you can’t make sense of the traffic, it’s too early to judge whether it’s reaching the right people or driving the right actions.

A better way to prioritize your AI search strategy is to move it through three stages: show up correctly, build and measure traffic, then qualify and convert it.

In other words: crawl, walk, then run.

The Crawl, Walk, Run Model for AI Search

Each stage of an AI search program answers a different business question.

1. Crawl: Are we present, and is what AI says about us accurate?

This is the foundation. Before optimizing for more AI visibility, you need to understand whether AI engines can access your content, where your brand appears, and how accurately your brand is represented.

2. Walk: Is our visibility growing, and how would we know?

Once the foundation is sound, the focus moves toward increasing presence and building enough measurement discipline to understand whether the program is gaining traction.

3. Run: Is this traffic worth having?

At the Run stage, the focus shifts to traffic quality and business impact. You’re looking at whether AI search is bringing the right people to your site, what they do once they arrive, and whether those visits contribute to meaningful outcomes. 

These stages are connected, but they’re not a rigid companywide ladder. An enterprise might be running in one product category while still crawling in another. What matters is correctly diagnosing each part of the business before deciding what to optimize next.

Crawl: Show Up Correctly

The question: Are we present, and is what AI says about us accurate?

The first stage of AI search is all about establishing a trustworthy foundation.

Start with technical accessibility. Can AI crawlers retrieve the content you expect them to find? Server logs can reveal crawler behavior that conventional analytics platforms cannot.

Not all AI crawlers do the same job, and the difference could change how you prioritize.

Training crawlers collect content for future model training. They shape what models know months from now. Retrieval crawlers build the indexes that AI answers get assembled from in real time. Those are the ones that determine whether you get cited this quarter.

If you have the access, it could be a helpful exercise to filter your server logs for these user agents.

Retrieval crawlers, which affect current citations:

• OAI-SearchBot: populates ChatGPT search
• ChatGPT-User: fetches pages when a user's prompt triggers a live visit
• Claude-SearchBot and Claude-User: Anthropic's equivalents
• PerplexityBot and Perplexity-User: Perplexity's index and live fetches
• Googlebot: still feeds AI Overviews and AI Mode

Training crawlers, which affect future model knowledge:

• GPTBot (OpenAI)
• ClaudeBot (Anthropic)
• Google-Extended (Google)
• Applebot-Extended (Apple)
• Meta-ExternalAgent (Meta)
• Amazonbot (Amazon)
• CCBot (Common Crawl)

If retrieval crawlers are not reaching your priority pages, nothing else in this framework will work. Check robots.txt first. A lot of sites blocked these agents in 2023 and 2024 and never revisited the decision.

Then look beyond branded prompts. A company searching its own name will naturally see a very different picture than a buyer asking a category-level question such as “What are the best ways to improve content performance?” or “Which platforms can help enterprise teams scale content?”

Those non-branded prompts reveal whether your brand is entering the conversations that precede consideration.

Presence alone isn’t enough, either. You also need to understand how AI search engines represent your brand. A citation may look like a win until you discover the answer describes your product incorrectly, relies on obsolete information, or reinforces a position you’ve spent years trying to change.

Finally, assess content hygiene. Crawlable markup, incomplete sitemaps, duplicate pages, outdated content, and pages competing against one another can all weaken the signals AI systems have to work with.

Done looks like: You know which prompts matter, whether your brand appears for them, and whether its representation is accurate.

Common mistake: Treating content volume as the objective. Publishing more content before fixing inaccurate or inconsistent AI representation can simply create more material reinforcing the wrong signals.

Typical owner: SEO or technical marketing.

Walk: Build and Attribute AI Search Traffic

The question: Is it working, and how would we know?

Once you understand your baseline, you can start expanding your presence across the prompts that matter.

That means monitoring whether visibility is improving, identifying gaps where competitors or other sources are consistently cited, and creating or updating content to strengthen your authority around important topics.

It also means building a measurement foundation.

Marketing teams can create AI channel groupings in analytics, monitor visibility metrics upstream of a website visit, examine relevant Search Console trends, and look at changes in traffic patterns over time.

Start with how AI traffic gets classified in your analytics platform. As of May 2026, GA4 has a native AI Assistant channel in the Default Channel Group. When it recognizes an AI referrer, it assigns a medium of ai-assistant and the session appears in that channel automatically. There is nothing to configure.

For most teams that should be the reporting baseline. It needs no maintenance, and it gives you a definition the rest of your organization will recognize. Treat it as a floor rather than a complete picture, for three reasons.

However, that source list may be incomplete. Google's documented list covers ChatGPT, Gemini, Copilot, DeepSeek, and Grok. Perplexity is not included and still lands in Referral. For B2B teams that is a real gap, because Perplexity traffic often shows stronger research intent than its volume suggests. Check Google's current documentation before assuming any given platform is covered. The list is still expanding.

It is not retroactive. Traffic from before the channel switched on in your property stays classified as Referral or Direct. Your AI traffic trend line effectively begins at the switchover date, and any year over year comparison spanning it is comparing two different definitions. Say that out loud in reporting instead of letting someone else find it.

It only sees referrers. Sessions that arrive without one, from in-app browsers, mobile apps, or a copied and pasted link, land in Direct regardless of how you define channels.

If you need Perplexity coverage, consistent classification across your full history, or a definition you control, build a custom channel group alongside the default rather than instead of it. Custom channel groups are a separate reporting lens, so running both does not double count. In Admin, open Channel groups, create a new group, add a channel called AI Search, and set Source matches regex. Move it above Referral in the rule order.

^(chatgpt\.com|chat\.openai\.com|claude\.ai|perplexity\.ai|www\.perplexity\.ai|gemini\.google\.com|copilot\.microsoft\.com|grok\.com|meta\.ai|you\.com|poe\.com|deepseek\.com)$

Review that list quarterly. The domains and how citations are passing referral data tend to change over time.

Expect one platform to dominate your early numbers either way. Conductor's research puts ChatGPT at roughly 87% of AI referral traffic, so a flat month for your program may just be a flat month for ChatGPT. Segment by source before you draw a conclusion.

From there, monitor visibility metrics upstream of the visit, watch Search Console trends, and track how traffic patterns shift over time.

What marketers cannot do is expect every AI interaction to resolve into one perfectly attributable click.

AI search happens across systems that expose different levels of referral information. Some interactions produce identifiable traffic. Others do not. That makes measurement less like reading one definitive dashboard and more like triangulating several signals into a defensible view of what is happening.

The objective at this stage is therefore not perfect attribution. It is measurement discipline.

Your team should know which numbers are directly measured, which are directional, and which require interpretation.

Done looks like: You can show whether AI search visibility and traffic are trending in the right direction, explain what the available signals mean, and distinguish measured performance from estimates.

Common mistake: Waiting for perfect attribution before progressing. That standard can keep an AI search program stuck indefinitely.

Typical owner: SEO working with analytics or marketing operations.

The harder question is how to build a credible measurement model when the click itself is incomplete. That deserves its own treatment, because AI search attribution requires a different approach than traditional organic search.

Run: Qualify and Convert AI Search Traffic

The question: Is this traffic worth having?

At this stage, the focus shifts from traffic volume to traffic quality. Teams begin examining whether people arriving through AI search resemble their ideal customers, which content they reach, and what happens after they arrive.

That requires looking beyond last-click conversion.

An AI search user may already have spent significant time researching a problem before clicking through to your site. They may arrive on a deep product or educational page, explore additional resources, return through another channel, or influence a buying process involving several other people.

A last-click-only view can miss much of that behavior.

Instead, examine engagement, landing-page patterns, assisted conversions, account quality, and downstream actions. Then feed what you learn back into your AI search strategy.

If certain prompt categories consistently produce highly engaged visitors from priority accounts, those topics may deserve greater investment. If visibility is growing around prompts that produce little relevant engagement, the prompt set itself may need to change.

This feedback loop is what moves AI search beyond an SEO initiative and into a broader demand and growth conversation.

Done looks like: You can say whether AI search is bringing the right people to your site and adjust your strategy based on what those visitors actually do.

Common mistake: Applying last-click conversion logic to AI search and concluding that valuable traffic is not working.

Typical owner: Demand generation or growth in partnership with SEO.

How to Diagnose Your Current AI Search Stage

As AI search programs mature, the challenge is knowing when you’re actually ready to move from one stage to the next. The questions below help you assess whether the foundation is in place at each stage, and where gaps could keep the rest of the program from working as intended. 

Stage

Self-Check

Crawl

• Do we know which non-branded prompts matter to our buyers? 
• Do we know whether we appear for them? Is our brand represented accurately? 
• Can AI crawlers access the content we want them to find?

Walk

• Are we growing visibility across the prompts that matter?
• Can we identify AI-referred traffic?
• Do we have a defined way to combine traffic, visibility, and search signals?
• Do we know which metrics are measured versus estimated?

Run

• Do we know whether AI-referred visitors match our ICP?
• Do we know which pages they land on and what they do next?
• Are we looking at assisted conversion and engagement, not just last-click?
• Are those insights changing the prompts and topics we prioritize?

For large enterprises, this assessment should happen by market, product line, or business unit rather than forcing the entire organization into one stage. One part of the business may be running while another is still establishing the fundamentals of crawl. 

The self-check tells you what to look at. It does not tell you when you are done. These are the bars I'd use.

Crawl to Walk: You have a defined set of at least 25 to 50 non-branded prompts tied to real buying questions. You have measured presence across all of them at least once. You have read the actual answer text, not just logged whether you were cited, and you have a list of the inaccuracies to inform of upcoming content strategy. Retrieval crawlers are reaching your priority pages, confirmed in server logs rather than assumed.

Walk to Run: You have at least a quarter of trend data on visibility and AI referral traffic. You can state a presence rate across your prompt set and say whether it moved, and can start to explain why, tying specific actions to visibility growth. AI referral traffic is separated in analytics and large enough to segment, which in practice means a few hundred sessions a month, not a few dozen. Your team can name which metrics are measured and which are estimated.

Running well: You can describe AI-referred visitors in the same terms you use for any other channel: who they are, what they do, what they are worth. Prompt priorities have changed at least once based on what that data showed.

Treat these as working bars, not industry standards. The right numbers depend on your category and your traffic volume. The discipline is having a bar at all, because without one every team grades itself as further along than it is.

In practice, that may mean separate GEO roadmaps for each part of the business that all contribute to one companywide strategy. 

What Comes Next

AI search programs will keep getting more sophisticated as measurement improves, platforms evolve, and teams learn more about how people use these channels.

The important part is knowing what your organization is ready to evaluate now. That keeps teams from overinvesting in metrics they cannot yet interpret or chasing outcomes they do not yet have the foundation to support.

From there, the work becomes easier to prioritize. Teams can focus resources on the gaps that are actually holding the program back, whether that is visibility, measurement, or understanding the quality of AI-driven traffic. As those gaps close, the next set of decisions becomes clearer, giving the program a more disciplined path forward.

Frequently Asked Questions

How do you improve AI search visibility?

Start by identifying the non-branded prompts your buyers are likely to use, then evaluate whether your brand appears in those answers and how accurately it is represented. You should also confirm that AI crawlers can access the content you want surfaced. Once that foundation is in place, you can focus on expanding visibility across the prompts and topics that matter most.

What should you measure in an AI search program?

The right metrics depend on the maturity of your program. Early on, focus on presence, citation, accuracy, and crawler accessibility. As the program develops, add AI referral traffic, visibility trends, engagement, landing-page behavior, assisted conversions, and account quality. The goal is to measure what you can reliably interpret at each stage.

When should you focus on AI search attribution?

Attribution becomes more useful once you have enough AI visibility and traffic to analyze. Before that point, teams are better served by establishing a reliable baseline and monitoring directional signals. Because AI platforms expose different levels of referral data, attribution typically requires combining traffic, visibility, analytics, and behavioral data rather than relying on one definitive metric.

How do you know when your AI search program is ready to scale?

You’re ready to scale your AI search program when the fundamentals of the current stage are working consistently. That means you understand where your brand appears, can measure whether visibility and traffic are improving, and have enough data to evaluate traffic quality and business impact. Different products, regions, or business units may reach that point at different times.

‍