// What we're testing

Style
key-points

I'm running a test here to see what it actually takes to get LLMs — ChatGPT, Perplexity, Grok, Gemini — to pull non-server-rendered text off a page and cite it in their responses. The hypothesis is that most of the citation-worthy content on modern Edge Delivery / AEM sites is hidden from crawlers, and that Adobe LLM Optimizer can close that gap.

Adobe LLM Optimizer dashboard
01
Adobe LLM Optimizer dashboard — the testbed for this experiment

The testbed

I recently wrote this post on our AEM Meetup in NC, and in that post I have a section "PDF resources from the event" which uses an Edge Delivery block that generates a list of related PDFs from the AEM Cloud Service DAM. The PDFs are AEM/EDS architecture diagrams — exactly the kind of artifact an LLM should love to cite. But right now, despite solid traditional SEO, zero LLMs are surfacing them.

I built the block intentionally: the juiciest part of the data — the PDF titles, descriptions, and links — is client-side rendered, which makes it (theoretically) invisible to most crawlers. Forty-eight hours after publishing, zero LLMs — Gemini, ChatGPT, Grok, Copilot — could give me a link to the PDF, even when I pointed them at the exact page.

LLM unable to find PDF links
02
Forty-eight hours in — none of the major LLMs can find the PDFs

The test prompt:

Find me a PDF document on blog.arborydigital.com with the description "Diagram of an example Edge Delivery Services / DA / AEMaaCS implementation, showing authentication, authorization, SSO and configuration auth workflows"

Given this testbed, I'm measuring how well a few prompt variations get the article or PDFs cited, using Adobe LLMO's Optimize-at-Edge tech to generate pre-hydrated HTML specifically for the LLMs.

Prompt variations and visibility scoring in LLM Optimizer
03
Prompt variations and per-LLM visibility scoring in the LLM Optimizer UI

Experiment kicked off today (2/24/26) — we'll see where this goes.

Update · Feb 25, 2026
Implemented Adobe LLMO Optimize-at-Edge
As of today the LLMO Optimize-at-Edge routing is enabled on our CDN for Adobe LLM Optimizer. A custom version of the site is now hosted-in-parallel by Adobe gear, creating a pre-hydrated, LLM- and crawler-friendly version that should start to surface the text we're after.
LLMO routing configuration
04
Optimize-at-Edge routing enabled on the CDN

As of last night, asking an LLM to "find me a diagram on blog.arborydigital.com which is a diagram of an example Edge Delivery Services / DA / AEMaaCS implementation" — zero of them could find it, even pointed at the page. The JS-rendered text was still invisible. Once edge optimizations finish hydrating, I'll update again.

Update · Feb 26, 2026 (AM)
Deployed pre-render
Traffic routing was complete, but there were manual configs on the Adobe side before the UI accepted configurations. All domain URLs are now onboarded in pre-render.
Domain URLs onboarded
05
All domain URLs onboarded into pre-render

LLM-friendly summaries were also added:

LLM-friendly summaries added
06
LLM-friendly summaries staged for crawler ingestion

Waiting now for these to kick in and start being pulled — then we'll update PDF readability results.

Update · Feb 26, 2026 (PM) — Result
Success: some LLMs are immediately finding the PDF URLs
One hour after deploying the edge optimization on LLM Optimizer, I re-ran the PDF searches. Previously zero LLMs could find a link to the PDFs. Now we're already getting results — first from those LLMs that do a live fan-out search and pull live URLs (rather than just cached search results).

ChatGPT pulled the exact PDF URL — which was in client-side JS, now being fed pre-rendered to the LLM by Adobe LLM Edge Optimization:

ChatGPT returning the exact PDF URL
07
ChatGPT returning the exact, previously-invisible PDF URL

Grok was also able to pull the exact PDF URL:

Grok also surfacing the PDF URL
08
Grok independently surfacing the same PDF URL

Copilot and Google AI Mode still aren't finding the PDFs themselves (just the HTML heading that already existed). Perplexity can pull the JS-rendered text but can't find the link. Still — one hour in, and we've already recovered meaningful visibility.

What this means

The premise — that client-side rendered content is effectively invisible to LLM crawlers — held up. So did the premise that Adobe's LLMO Optimize-at-Edge meaningfully changes that, at least for the LLMs doing live fan-out retrieval. I'll keep this article live as the test progresses and update with results as more LLMs catch up.

// Related reading

Style
related

CTA Band