GPTBot visits every 2 days. ClaudeBot every week. Perplexity only when asked.

Three AI crawlers showed up at your site this morning. They left with three different things, on three different schedules, looking for three different signals. If your AI search strategy treats them as one channel, you’re optimizing for something none of them actually do.

A 30-day server log study published April 26, 2026 by Digital Applied tracked 11 AI user-agents across 12 production sites. The numbers are blunt. GPTBot averaged 4,200 hits per site per day, ClaudeBot 1,800, and PerplexityBot 980. The frequencies aren’t the interesting part. The shapes are.

GPTBot crawls breadth-first, hits your /blog/, /docs/, and /about/ paths, and revisits high-traffic pages every 2.4 days. ClaudeBot crawls depth-first, averaging 5.2 levels deep against GPTBot’s 3.8, and prefers technical documentation. PerplexityBot doesn’t crawl in the conventional sense. It fetches on demand when a user queries a domain, and when it does fetch, it bursts at 200+ requests per minute.

Each bot is reading your site for a different reason. Optimizing for one without measuring the others is how teams end up writing for an audience that never visits.

The case for crawler logs as the truth source

A year ago, at SEO Week NYC in April 2025, Jori Ford put a name on the work most teams aren’t doing. She called it Hybrid Engine Optimization, and she made one argument worth taking seriously. Server access logs are the ground-truth signal for AI visibility, not whatever a SaaS visibility tool tells you about your “share of voice.”

The point wasn’t the acronym. The industry didn’t need a fifth one. SEO, AEO, GEO, LLMO, and HEO is more frameworks than actual differences in tactics. The point was the data source. Most GEO platforms tell you which queries cite your brand. They don’t tell you which AI crawlers visited the page that earned the citation, how often, or what they looked at while they were there. Visibility tools sample. Server logs are exhaustive.

Two things follow. First, if your team has been running an AI visibility program without log analysis, you’re working with a downstream signal. The tools see effects. The logs see causes. Second, the recent shift in crawler behavior has made log analysis more useful, not less. GPTBot and ClaudeBot both started consuming sitemap.xml for the first time in March 2026. If you don’t know whether they’re hitting yours, you don’t know whether they’re seeing your new content.

What each crawler actually does

The table below summarizes the crawler-level behavior from the study. The differences matter because they imply different optimization priorities for each platform’s downstream answer engine.

Crawler	Hits/day	Revisit cadence	Crawl shape	What it cares about
GPTBot	~4,200	2.4 days on top pages	Breadth-first, 3.8 levels avg	/blog/, /docs/, /about/
ClaudeBot	~1,800	6.8 days	Depth-first, 5.2 levels avg	Technical documentation
PerplexityBot	~980	On-demand	Burst, 200+ req/min	Pages users query
OAI-SearchBot	~600	Daily re-fetches	Targeted	Cited URLs

The behavior tells you what each downstream engine actually has to work with. ChatGPT’s answer engine is consuming a recent, broad sweep of your content. Anything more than 48 hours old is in the index. Claude’s engine has older, deeper material. A post from last week is likely in scope. A post from yesterday is probably not. Perplexity has whatever the user’s query forced it to fetch, which means a brand new page can land in a Perplexity answer the day it ships, but only if someone queries a topic that triggers a fetch.

That last detail is the one most teams misread. Perplexity is the only major engine where you can move the needle by publishing content the same week. It’s also the only one where freshness is structurally rewarded over authority, because the engine isn’t ranking from a cached index. It’s pulling fresh.

Where the acronym debate gets in the way

Most teams I talk to are stuck in a metadata fight. Are we doing GEO or AEO? Should we add LLMO to the strategy doc? Does HEO replace the others? The fight is a distraction. The work doesn’t change. You audit your crawler logs, you find the bots actually visiting, you check whether they’re reaching the pages you want them to reach, and you fix whatever’s broken.

The acronym proliferation is also a tell. Five frameworks in eighteen months is what an industry produces when nobody is sure which signal is load-bearing. Once the answer engines stabilize on a set of behaviors, and the crawler data above suggests they’re starting to, the industry will collapse the frameworks back down to one or two. Until then, ignore the labels and read the logs.

What this changes about your optimization plan

The crawler data implies a few specific shifts in how teams should allocate effort. We covered the fundamentals of robots.txt strategy for AI bots earlier this year. The new data sharpens what you’re allowing in for.

1. Audit your sitemap. Both GPTBot and ClaudeBot started reading sitemap.xml in March 2026. If yours is stale, two of the most active AI crawlers are getting an outdated map of your site. If your CMS doesn’t refresh sitemaps on publish, fix that first.

2. Move technical content into shallow paths. ClaudeBot averages 5.2 levels deep, but GPTBot averages 3.8. If your best technical content lives at /learn/category/sub/article, GPTBot may never reach it. The same article at /docs/article lands in both crawlers. Path depth is a discoverability lever now, not a navigation choice.

3. Use Perplexity’s on-demand behavior as a launch tool. Because PerplexityBot fetches on user query, you can push fresh content into Perplexity’s answer pipeline within days by seeding the relevant queries. We wrote about how content placement drives AI citations, and the same logic applies here with even tighter feedback.

4. Stop running “AI visibility” as one program. The crawler data is the clearest argument that ChatGPT, Claude, and Perplexity are three separate distribution channels with three different read patterns. Citation overlap between them is already low. Only 11% of websites are cited by both ChatGPT and Perplexity. The crawler behavior is why.

The gap between what tools measure and what bots do

Most AI visibility platforms, including ours, sample query results to estimate brand presence. That’s the right metric for outcomes. It’s the wrong metric for diagnosis. When a brand drops in citation share on ChatGPT, the platform tells you the result. The logs tell you whether GPTBot stopped crawling a critical path, whether your sitemap broke, whether a recent CDN change started serving 403s to OAI-SearchBot. Those are all causes the result-level data can’t see.

The teams winning AI visibility in 2026 are running both signals together. Outcome data from a visibility tool. Cause data from server logs. The work isn’t to pick a framework. It’s to make sure the two signals line up.

If they don’t, you’re optimizing for what the tool tells you, not what the bots are doing. That gap is where most AI visibility budgets get burned.

RivalHound tracks your brand’s visibility across ChatGPT, Google AI, Perplexity, and more. Start monitoring to see where you stand.