A simple test every brand should run today and what to fix once you've run it. Including how titles and slugs decide whether ChatGPT, Rufus and Gemini cite you in their answers.
A simple test every brand should run today and what to fix once you've run it.
Most brands think they know what their website looks like.
But what matters in 2026 is what LLMs can see - not what your browser shows you after all the scripts have run.
If your content is hard for models like ChatGPT, Claude, Rufus, Gemini and Perplexity to read, you are invisible at the exact moment shoppers are asking questions that should lead to your product.
Update — May 2026: Amazon has merged Rufus with Alexa+ to create Alexa for Shopping, now live on the Amazon Shopping app, website and Echo Show. References to "Amazon Rufus" in this post relate to the predecessor product. Read Amazon's announcement.
And the surprising bit:
A huge chunk of modern websites are very hard for LLMs to read in a reliable way.
Here's a simple way to see how dependent you are on JavaScript and what to do once you know.
The one-minute sanity check
This is not a perfect simulation of how every crawler works. But it's a fast way to see how much of your site exists before JavaScript runs.
- Open your site in Chrome
- Right-click → Inspect (or Cmd+Option+I on Mac, Ctrl+Shift+I on Windows) to open DevTools
- With DevTools open, press the Command Menu shortcut (Cmd+Shift+P on Mac, Ctrl+Shift+P on Windows)
- A search box appears at the top of DevTools. Type: Disable JavaScript
- Click the "Disable JavaScript" option that appears
- Keep DevTools open and refresh the page (Cmd+R / Ctrl+R)
What loads next shows you how much of your page is there without JavaScript.
If a lot disappears, you are heavily JS-dependent.
That is your first signal.
Why this test matters (and what it doesn't tell you)
Most LLMs lean on HTML and structured data. Many do not fully execute every script on every page they ingest.
So if your product information only appears after a JS framework renders it, or your PDP content is loaded via async calls, or your "experience layer" is a shell without JS - simpler crawlers and some LLM pipelines will see a much thinner version of your site than your customers do.
The "disable JavaScript" view is a diagnostic, not the full truth. LLMs can still learn about you from your raw HTML, JSON-LD structured data, sitemaps, feeds and from retailers, marketplaces and review sites that carry your data.
So if the page looks empty with JS off, don't panic. Treat it as a prompt to check what your HTML and structured data are actually exposing.
Step two: check what the HTML really contains
Pick one key page. A product detail page or a category page works best.
Load it normally with JavaScript enabled. Right-click and choose View Page Source.
In the source, look for:
- Real text content - product titles, descriptions, features, headings, bullet points. Not just placeholders waiting for scripts to fill them.
- Structured data blocks - anything inside
<script type="application/ld+json">. - Meta tags that describe the page - title, description, og:title, og:description.
- Heading hierarchy - h1, h2, h3 in the right order.
If you can read the page's substance in the source view, an LLM probably can too. If you only see a wireframe of div tags and skeleton placeholders, the model is reading the same wireframe.
This isn't always a fix-it-tomorrow problem. But it's a "you need to know" problem. Brands that ship server-rendered or hybrid-rendered pages have a structural advantage in AI search that fully client-rendered SPAs don't.
Step three: test your structured data
JSON-LD is the part of your HTML that talks directly to machines. It's also the part most brands either skip or get wrong.
The fast check: paste any product URL into Google's Rich Results Test and Schema.org's validator. You're looking for three things.
- Is there a Product schema block at all? If no, that's the first job.
- Are the fields complete - name, brand, sku, gtin, price, availability, image, description, aggregateRating?
- Do the values in the schema match the values on the visible page? Mismatches reduce trust. Models that triangulate signals between the visible page and the structured data will downgrade pages where the two don't agree.
For consumer brands, also ship FAQPage schema on PDPs and category pages and Organization schema on the homepage with sameAs links to your social and retail profiles. That's the minimum.
Update - May 2026: Google deprecated FAQ rich results on 7 May 2026.
FAQPageschema no longer earns a visible feature in Google Search. It still has value for AI assistants - ChatGPT, Claude and Perplexity use it to parse Q&A content - but you do not need it for a Google rich result.
If structured data is unfamiliar territory, our foundational guide to Schema.org for LLM optimisation walks through which types matter, why they matter and what to ship first.
Once you're readable, get cited: match how shoppers ask
Being readable is the floor. Being cited in the answer is the ceiling. The two require different work.
A recent AirOps analysis of snippet signals makes the point clearly: AI engines decide which page to cite using a mix of intent matching and language precision. The patterns split by query type.
For informational queries - "what is", "how to", "best practices" - precision is everything. The AI engine prefers titles and slugs that echo the exact query language. A page titled "What is Amazon Rufus?" beats "Understanding the New Amazon AI Shopping Assistant" for the query "what is Amazon Rufus", even if the second post is longer and better written. Mirror the question.
For commercial queries - "best running shoes for marathon trainees", "vegan SPF50 sunscreen for sport" - the engines value semantic breadth more than literal match. Use synonyms, related categories and the language shoppers actually use. A page that only says "high-protection sunscreen" misses the prompts that say "SPF50 sport sunscreen" or "sweat-resistant sunblock".
The simple rule: write title and slug to mirror informational prompts and write body content to span commercial-prompt vocabulary.
Two tactical moves you can ship this week:
Audit your top 30 page titles against the actual buyer prompts they should rank for. If your titles read like internal taxonomy ("Sunscreen Range", "Our Hydration Products"), rewrite them as the question or the comparison shoppers type. "Best SPF50 sunscreen for runners" beats "Sport SPF50 - Our Range".
Build comparison and "best of" pages that use shopper language in headings and the first paragraph. Models extract from the top of the page first. Make the lift easy.
What to ship this week
Three things. In order.
Run the disable-JavaScript test on five pages - homepage, top category, top PDP, a key explainer post and your About page. Note what disappears.
Validate JSON-LD on those same five pages. Fix any missing Product, FAQPage or Organization blocks.
Rewrite the top ten page titles so they mirror the actual buyer prompts. Don't over-think it. The prompt is the title.
If you do all three, you've moved from "the model can't really read me" to "the model reads me and knows how to cite me". That's the gap most consumer brands have right now.
That is the shift.
If you want a Lmo7 LLM-readability audit on your site, including the JSON-LD validation and the title-prompt mapping, get in touch.