The cost of being invisible to AI. A SaaS pricing case study.

Consider a mid-market B2B SaaS: organic traffic flat, but pipeline coming in under forecast. Google Search Console showed no rank changes. The content team was shipping on cadence. Conversion rates on the pages that DID get visited were stable. The numbers all looked fine, and the pipeline was still soft.

This is an illustrative composite drawn from the patterns we see across audits, not a single named account. It's a story about what a team in this position tends to find and what they can do about it.

#What we found

We ran a Crawlmind audit against their site and then ran their top 50 buyer-intent queries through ChatGPT (with browsing), Perplexity, and Gemini, logging citations.

Three findings:

#1. Their site appeared in zero of the 50 AI citation sets

Not 5, not 2. Zero. For every query a target buyer would actually type, the AI engines cited competitors, review sites, and Reddit threads. The customer's own thought-leadership pages weren't even in the consideration set.

This was striking because the same queries returned their pages in Google's top results for most of them. The gap between "ranks well in Google" and "gets cited by AI" was almost total.

#2. The technical causes were knowable

The audit flagged:

• No llms.txt
• No author attribution on blog content (the byline was "Marketing Team" for everything)
• Three of their highest-traffic pages had dateModified from 2022
• Their robots.txt blocked GPTBot, PerplexityBot, AND ClaudeBot (added 18 months earlier by a security-conscious engineer who'd read a blog post about "blocking AI scrapers")
• No Article schema on any post, just generic WebPage

Individually, each of these is a small thing. Cumulatively, they made the site close to invisible to AI engines while perfectly visible to Google.

#3. The pipeline impact was real

We did the back-of-envelope on attribution. Their highest-converting discovery channel had historically been "organic search." Of that, a meaningful chunk appeared to come from long-tail comparison queries that AI engines were now answering directly. If that much of the organic-discovery funnel was being summarized away by AI, the pipeline softness was directionally consistent.

This is the worst kind of metric problem: the leading indicators (impressions, rank, traffic) don't show it, the lagging indicator (pipeline) takes a quarter to materialize, and the cause is invisible to standard SEO tooling.

#What they did

A six-week sprint, in roughly this order:

Week 1. Unblocked GPTBot, PerplexityBot, ClaudeBot in robots.txt. This alone took 30 seconds.

Week 2. Shipped a proper llms.txt with sections for "Core guides," "Product," and "Optional resources." Linked their top 25 canonical pages. Added a CI check that breaks the build if any link in llms.txt 404s.

Week 3. Added BlogPosting schema to all blog posts with proper author (replacing "Marketing Team" with the actual writer's name linked to a new author page), datePublished, and dateModified.

Week 4. Refreshed the top 20 pages by content team. Real edits, not date-bumping. New stats, updated examples, removed dead links. Pushed live with current dateModified.

Week 5. Built author pages for the four most-active writers with LinkedIn profile links and short bios. Added Person schema.

Week 6. Set up weekly citation tracking via Crawlmind on the top 50 queries.

#What changed

In a scenario like this, citations climbed from none to a handful within weeks, and to a meaningful share of tracked queries by the end of the quarter, then leveled off:

• The first pages to get cited are specifically the ones refreshed with real edits, not the ones where only the date was updated
• Perplexity citation share tends to start moving first, then ChatGPT
• Direct traffic to the cited pages increases modestly, and the share of trials citing "discovered via AI" on the signup form (added as a signup question) rises noticeably
• Pipeline forecast comes back on track (causally fuzzy, but directionally consistent)

#The lesson

The work wasn't dramatic. Unblocking three crawlers, fixing some schema, writing better blog post leads, refreshing dates with real edits. None of it was novel. None of it was expensive. The hard part was knowing to do it, and the hard part of that was knowing it mattered, and the hard part of THAT was having a way to measure the gap in the first place.

The metric that mattered (citation rate by query) didn't exist in their analytics stack. They were running blind, and the first signal was a pipeline miss. Don't be them. If you sell to people who use ChatGPT, measure your AI citations directly. Whether with Crawlmind or with a spreadsheet and a habit of running the same 20 queries every Monday.