AI Search for Non-Developers
Cutting through the confusion of “AI Search”
Firstly, before people switch off, this isn’t an article trying to sell you a new AI tool! It’s about understanding a technology well enough to make better user experience decisions.
Over the past few years, I’ve been asked many times about the benefits of adding AI search to websites. The problem is that there’s so much confusion about what AI search is, that it’s too easy to get bogged down in all the different options and terminology.
I lead the design team at Liquid Light, working primarily with NGOs and other large organisations who often have large knowledge repositories. I’m a designer at heart, not a developer. But I’ve learned that some technical knowledge (even at a high level) goes a long way in helping us make better informed decisions about serving our audience.
So if you want to know what terms like AI search, semantic search, contextual search or vector search mean in a non-techy way, this article is for you. Whether you are considering upgrades to your website, evaluating a proposal from your agency or just trying to understand what your team is recommending, these distinctions make all the difference.
Breaking down the terminology
Semantic search
This one is easiest to explain by comparing it with a “keyword search”:
- You type “plastic pollution”
- Results show pages with “plastic” AND “pollution” (or just one of those words, depending on settings)
The issue here is that unless those exact keywords are actually on the page, the search won’t pick them up.
A solution to this is to use a search engine that understands the meaning behind words (“semantic search”):
- Your users search for “plastic pollution”
- Your content includes “marine debris” or “microplastic contamination” or “ocean waste”
- Your search understands that these are to do with plastic pollution and gives you results
This is where the AI comes in. It’s the AI models that are used to learn the relationship between words.
Vector search
This is the “how” behind semantic search. The AI models convert words into numbers so that they can calculate that “plastic pollution” is mathematically similar to “marine debris” for example. These numbers are called “vectors”. Vector search IS semantic search, it’s just the technical term.
Contextual search
Contextual search considers who’s searching, where they are, the time and date and what they have searched for previously. Google does this very well where it will show me different results to you for the same terms.
Plain/Natural language search
Keyword search may struggle with something like “Are microplastics in the ocean there forever?”. It looks for pages with all those keywords and gets confused. Semantic search handles this better, it understands the question you are asking. It’s not a separate search technology, it’s a feature of semantic search.
Hybrid search
This is a combination of keyword search and semantic search. Here the search algorithm will use both, creating a combined list of search results. Why is this useful? Because each has different strengths. Consider this example:
If someone searches for “ISO 14001:2015” they are looking for something relating to that exact standard (keywords works better). If someone is searching for “environmental management certification”, they are looking for a conceptual understanding (semantic is better). Hybrid allows for the precision of keywords with the understanding of semantics.
Generative search
When someone asks us about AI Search, often they are really picturing “adding ChatGPT to our website”. Everything I’ve talked about so far (semantic search, hybrid search etc), uses “Embedding models” and not Large Language Models (LLMs). Embedding Models understand meaning, but they don’t generate text. ChatGPT (and Claude, Gemini, Copilot or your service of choice) all use LLMs.
This matters because when LLMs generate text, they do it by predicting the next most likely word which is why they can confidently make things up (hallucinate). Generative search is optional. You can have all the benefits of semantic search without the downsides of AI generated text, which is often the best choice when accuracy is critical.
Why this matters
So what actually is AI Search?
Based on the above you can see that AI search could mean many things. For example it could mean semantic search, generative search, contextual search or a combination. The issue with the term AI Search is that it’s become a bit of an umbrella marketing buzzword, when it’s really a combination of different technologies. When terms become broad like this, there’s a risk of adding technology to websites without proper consideration.
Why now?
AI search has been around for years, but it’s only recently that the cost for implementing semantic search has dropped to an accessible level. At the same time, services like ChatGPT and the usage of AI in Google’s search results have raised expectations. For organisations with large knowledge repositories (research papers, policy docs, training materials) this gap between expectation and site reality can be frustrating.
Interestingly, at the time of writing, the vast majority of content management systems (for example Wordpress, Drupal, Joomla, TYPO3), don’t support AI search natively (without extensions or 3rd party integrations). This means that organisations must make using the technology an active decision, not just switching on an inbuilt feature.
What to Consider
When evaluating AI search, it is important to ask specific questions such as:
- Would generative search be helpful? Does the trade-off over hallucinations outweigh benefits?
- Does semantic search solve our specific problem, or would keyword search give us more focused results?
- Does our content need to be organised better so there is less reliance on search?
- Will the additional costs (financial and environmental) justify the improvements?
That last question is particularly important to many of our clients. AI services have significant energy costs, and for organisations committed to sustainability this isn’t a small consideration.
A real world example is that we’re in the process of improving the search functionality for a client with different audiences. Each audience uses different search terms to find content. In their sector a lack of accuracy can cost money through litigation. For them we’ve used a semantic search but avoided the hallucination risks of generative search.
Making the right decision
The role of anyone responsible for a website is to improve the experience of users. It’s not to implement a technology because it’s trendy. For search, that means knowing whether AI Search genuinely solves a problem your users are experiencing. Understanding these distinctions will help you evaluate what is being proposed and decide what will truly serve your audience.
-
Matt Keogh
Creative Director and Co-owner
