AI search KPIs that measure citations and share of voice can tell you whether your brand appears in AI answers. What they rarely tell you is what to fix next, what content to build, or where to shift budget. A framework presented in an on-demand Search Engine Journal webinar maps four first-party signals that can add a performance layer to the benchmarks most teams already use.
Why do citation counts fall short as performance signals?
Most AI visibility tools work by simulating prompts and tracking how often a brand gets mentioned or cited. Stas Levitan, Founder of LightSite AI, argues in the SEJ webinar that these measurements are useful for competitive benchmarking and trend analysis, but they are often asked to carry more weight than they should.
AI answers are contextual, personalized, and variable. A prompt sample can show what might happen. First-party site data shows what actually happened. Levitan calls the risk of conflating the two “the benchmark trap”: false confidence that pushes budgets toward activity that does not produce demand.
Citation counts can also fluctuate without giving marketers a clear next step. A spike in citations this month does not automatically point to the page that drove it, or explain why a previously visible page dropped off.
What are the four AI search KPIs that guide real decisions?
The webinar maps AI search performance across four stages: machine discovery, machine interest, human demand, and the ratio between them. Each signal answers a different question about what is happening on your site.
- AI bot traffic: Visits from AI crawlers to your pages. A top-of-funnel discovery signal showing that an AI system has accessed a page. Levitan treats bot visits as a new kind of site-level impression, a proxy for machine awareness before any human is involved.
- Pages consumed by AI bots: Which specific resources bots fetch, ignore, revisit, or explore more deeply. Page-consumption patterns can expose which content AI systems treat as worth returning to, versus content they skip.
- Human visits from AI search: Referral traffic indicating that an AI answer led a person to click through to your site. This signal bridges machine attention and actual human demand.
- AI click-through rate (AI CTR): The ratio comparing machine attention with human demand at the page, site, or campaign level. Levitan pairs this signal with a decision matrix that maps each outcome to a practical next action.
These signals are not interchangeable, and a high number in one area does not automatically guarantee an outcome in another. The framework is designed to be read together, not in isolation.
Why is bot attention concentrated on so few pages?
One of the more striking findings in the LightSite AI dataset: roughly 12% of pages absorbed about half of all bot impressions across the hundreds of websites studied, according to the SEJ webinar. A smaller group of pages that were re-read across a four- to six-week window accounted for a disproportionate share of total crawl volume.
The data also shows a pattern in the type of content that earns both bot attention and human visits. Tools, templates, support resources, and pages that answer one specific question for one specific audience consistently outperformed generic blog content on both dimensions. Frequently crawled pages did not always earn human visits, and vice versa, which is exactly where the AI CTR signal becomes useful.
What blocks AI bots before any of this can work?
The webinar opens its action framework with a technical check: confirm that major AI bots can actually access the site. Levitan reports that about one-third of websites in the dataset blocked at least one major AI bot, often not by intention. Security tools, CDN (content delivery network, the infrastructure layer that routes traffic) configurations, and bot-management settings made by different teams can inadvertently block AI crawlers even when the marketing team assumes the site is fully open.
CMS platform choice matters less than these infrastructure layers. WordPress, Shopify, Webflow, or Squarespace all have similar base accessibility, but CDN and security settings can override any of them.
How can marketers turn these signals into a content plan?
The framework Levitan outlines starts with access, then moves to prioritization based on observed signal patterns rather than fresh content creation. Before commissioning new content, the webinar recommends checking:
- Whether major AI bots can access your site without being blocked by WAF rules (firewall filters that screen incoming traffic) or CDN settings.
- Which pages are already receiving the most AI bot traffic, using server logs or first-party analytics.
- Whether those high-bot-traffic pages are also generating human referrals from AI search, or whether a gap exists.
- Which pages are being revisited by bots across multiple weeks, since repeated attention may signal higher perceived value to the AI system.
- Whether the highest-intent support, comparison, or answer pages are focused on a single specific question rather than covering broad topics.
- What the AI CTR ratio looks like at the page level, to identify pages with strong machine attention but weak human follow-through.
- Whether the teams managing security, CDN, and marketing are aligned on bot-access policy.
What this means for AI-search visibility
Here is the clearest way to think about what Levitan’s framework does: it separates “being seen by AI” from “being chosen by AI,” and adds a third question: “does that choice lead a real person to your site?”
Think of it like a paid-search funnel most marketers already know. Impressions (AI bot visits) tell you that you showed up. Clicks (human AI referrals) tell you that someone acted on it. CTR tells you whether the match between your content and the AI’s judgment is working. Most teams currently only track the impressions equivalent, and they are trying to optimize from there.
The concentration finding deserves close attention. If 12% of pages absorb roughly half of all bot impressions in a typical site dataset, per LightSite AI’s webinar data, AI systems are already making editorial judgments about which content matters on your site. In Hingewise’s view, this changes the practical question. It is no longer “how do I create more content?” It is “do I know which pages are already in that 12%, and are they actually good enough to earn a click?”
The access-blocking finding is underappreciated. A bot that cannot reach your site cannot cite you, regardless of how authoritative your content is. The fact that roughly one in three sites in the dataset had an unintentional block suggests this is a widespread, largely invisible problem, one that sits entirely outside the content team’s line of sight. It is often a conversation that has never happened between the security team and the marketing team.
Watch for whether major AI search platforms begin publishing more transparency about which content structures they consistently surface. That signal, combined with first-party crawl data, would significantly sharpen the kind of decision matrix Levitan outlines here.
Lam Nguyen, Hingewise
Sources: Search Engine Journal, New AI Search & SEO KPIs: 4 Signals That Guide Real Decisions (Stas Levitan, LightSite AI; Loren Baker, SEJ)
This article was drafted with AI assistance from the sources cited above and checked against them before publishing. The "What this means for AI-search visibility" section is Hingewise's editorial analysis, built on the frameworks and positions we use in our GEO work. How we work: Editorial policy.