Getting Website Content Into ChatGPT

In addition to testing our own content, we crawled the content of about a dozen websites in our main areas of focus — B2B, Healthcare, Higher Education, and Nonprofit — and fed the content into ChatGPT so we could ask it questions.

Try It Out: Ask our AI search a question below!

Use the interface below to ask questions about us and get results based on our website content. You might ask, “Do you create WordPress websites?” or “Tell me about your SEO services”.

Our Top 5 Takeaways From This Exercise

1. Generative AI doesn’t just ingest keywords; it interprets content

This one is probably obvious, but ChatGPT tries to “understand” content, not just store it and repeat it back. As a result, it will return answers that don’t match your exact website copy. It references the website copy, but its answers are only based on what it found. In our tests, we found that adjusting the prompt that we give to ChatGPT about how to respond was important in getting it to produce the kind of results we want. However, it also identified gaps in our content strategy. If ChatGPT can’t understand and infer the right thing from your content, search engines might not be able to either.

What it means for SEO: Continue to create clear, complete, well-structured content that accounts for some of the other nuances we point out below.

2. It understands prose best

While we found that ChatGPT was able to understand webpage content as it was structured well, it responds best to prose. It’s great at interpreting the sentence “The generative AI conference will begin at 9am on Saturday November 15th and the keynote will begin at 9:15am”, but may not always perfectly interpret the following:

Generative AI conference

Saturday November 15
9am
Keynote to begin at 9:15

What it means for SEO: In our tests, ChatGPT is good enough at interpreting non-prose website content (like the layout above), so we’re not recommending that you change all of your nicely laid out content at the expense of the user.

3. It can be heavily influenced by the wrong things

In a test on a hospital system website, ChatGPT kept responding that the hospital didn’t offer medical services, which was obviously not accurate.

After some troubleshooting, we determined that ChatGPT was taking the language of their legal disclaimer very seriously. The disclaimer noted that nothing on the website should be construed as medical advice and that the website itself should not serve as an alternative to medical services.

What it means for SEO: Our assumption is that Google and other search engines will tune the algorithm to control for this sort of scenario, but it has made us consider blocking legal terms and conditions from robots.txt since we generally don’t want to send organic traffic there anyway.

4. Old content can confuse it

We came across an interesting scenario where, for a company with many locations, we asked if there was a location in Pittsburgh.

The company does have a location in Pittsburgh, but the GPT response said that the Pittsburgh location would be opening in the fall. Why? Because there was an old press release stating just that: The Pittsburgh location would be open in the fall.

The only other mention of Pittsburgh was on the “Locations” page that just had the locations listed as cards. Since there was no content on the website that stated “The Pittsburgh location is now open,” ChatGPT was relying on the outdated information.

What it means for SEO: Clean up old content and, in instances where you can’t write prose, make sure the headings, structure, and description of the page give clear, text-based indications as to what’s on the page, what the hierarchy is, and how related data is grouped. Consider creating follow-up news and press releases if you announce that something is on the horizon.

5. ChatGPT can be moody

There are settings within the ChatGPT interface that tell it how “fuzzy” an answer to provide (i.e., whether to read the content very literally or rely more on its own language interpretation), and its responses are heavily influenced by how it is prompted to respond.

We got the best results when we told ChatGPT to act as an “honest, informed representative” and explicitly told it “do not make up an answer if you don’t know one”. Otherwise, it can start to speculate too much and give plausible but not entirely accurate responses.

What it means for SEO: Google and other searches engines are likely learning to tune these types of parameters, so expect even more changes and core updates that will continue to impact how your content is indexed and interpreted. (Google performed several major algorithm updates in 2023.)