Thoughts

Site Search vs. AI Assistant: How to Choose Your Search Tool

Author

Corey Vilhauer

Categories:

A graphic depicting a robot icon with a headset, representing an AI search bot, and a magnifying glass icon, representing traditional site search.

The right site search solution depends on what your visitors actually want — and whether your content is ready to support it. Here's how to decide.

We’re working on two low-stakes house projects right now — replacing a toilet, and adding new blinds to our patio door. Which means we need: a toilet, and some blinds.

One of these was simple. I went to Lowe’s and I asked where the toilets were. They pointed me to the toilets, and I picked out a toilet. I knew what I wanted, but I needed to know how to get there.

The second was more complex. We needed to walk through options — styles, cost, installation — so we had a conversation with someone who specializes in banded blinds. We gathered information on how to install them, what models worked best, and even how much they might weigh. We left with information that we then used to make a decision a few days later.

Both of these could have gone differently. If I’d asked where the toilets were and the Lowe’s employee didn’t know where they were, or if they were placed in the wrong aisle, I’d be very frustrated. Likewise, if I’d encountered an employee with no understanding of the ins-and-outs of banded blinds, I’d be left with mixed messages and less certainty.

I am a human, who was searching for things, so I talked to other humans. But this is essentially how site search works. We have different options for how to handle and improve our on-site search, and not every method works for every search. Let’s jump in and spell it out.

Understanding the reason behind site search.

There’s a common assumption that fixing site search means finding a new tool. This is what every vendor would like you to believe — that a plug-in or content management system (CMS) add-on is all you need to make search work a lot better.

In most cases, this is half true. The tooling matters, very much! There are certain features and common functionality that helps improve the usability and maintenance of site search. But site search is only as good as the content it's searching — and the way that content is structured, indexed, and ranked. Which means fixing site search is a question of what’s wrong with site search in the first place.

To understand this, we need to know what visitors want.

The reflex to treat site search as a shopping decision skips our understanding of how visitors are using our site search. Before you compare tools, you have to know what expectations are present around the search box. Different situations need different tools.

It can be as simple as answering two questions:

  • Do your visitors already know what they want, or are they asking open questions?
  • Do your visitors need to land on a page, or do they need an answer or to finish a task?

The answer isn’t a clean decision for any site. For example, a person looking for the undergraduate admissions application isn’t asking for a chat or researching an issue — they are looking for a specific thing that lives in a specific place. On the other hand, a future student wondering which programs fit their specific situation wants an answer that pulls from several pages. These are different jobs, and the tool that serves one can be the wrong tool for the other.

No one is happy with an “It Depends” answer, so do your own research — sort your internal search traffic against these two questions and determine which is most likely for your site.

When better site search is the right move.

Good site search is complex. It is complex because we expect more out of it than we end up getting. We have spent a decade using advanced search tools like Google, which clouds our thoughts on on-site search — we expect Google-like results, despite not having Google-like resources.

Traditional site search is primarily a retrieval service: it finds and ranks the pieces of content or records that best match a query. Modern search can go well beyond exact keyword matching, using synonyms, semantic ranking, vector search, and other techniques to understand what a visitor means. But the core goal of traditional site search is to search for a page you know is there and then go to the page you know is there.

For this kind of work, an AI assistant may be unnecessary. Instead, you need a search tool that has been tuned for the words that people use. This means synonyms, and “elevated” results tied to unique terms, and filters that match how people narrow down results. This might even mean federated search if your content lives across multiple sites.

This is also relatively simple to understand, because it depends on using the right terms. The downside of traditional site search is also that it depends on using the right terms. The first time you visit a banking site, you probably assume there’s a page for checking accounts. The first time you visit a hospital site, you probably assume there’s a page with parking maps. This is where the frustration can set in — your assumption might not match the reality, for a few reasons:

  • If the page is not there, it cannot be surfaced.
  • If the page uses a term different from what you typed, it might not be surfaced.
  • If the answer is trapped inside a PDF or buried in a page about something else, it might not be surfaced.
  • If the page is present, but out of date, search will surface it anyway — wrong information and all.

Traditional site search is an unglamorous option, but it’s often a more affordable option and, if your site is not content rich or complex, it’s also a more sensible option.

A set of colored lines turning into a bundled mess, signifying technical debt.

What is organizational AI readiness?

An Organizational AI Readiness Assessment helps organizations get a clear, honest picture of where they stand with AI — what's already happening, what opportunities are worth pursuing, what risks need to be managed, and what a reasonable path forward looks like.

Learn more with our organizational AI readiness explainer.

 

What is an Organizational AI Readiness Assessment?

When an AI assistant is the right move.

On the other hand, there’s the other kind of query: an open question that may not necessarily map to a single page.

Open questions are rarely answered by a list of links. A traditional search engine results page, in this case, is akin to assigning homework to the search user — they ask a question, and you give them a list of resources to pore through on their own. This is a strong use case for an AI assistant, because it can synthesize information from multiple relevant pieces of content into an answer.

This is even more so when the end state of a search is not a page at all. If a parent wants to find an appointment to meet with their child’s doctor, the end state is not to read a page about appointments, but to make the appointment itself. AI assistants can be integrated with those underlying systems to streamline this process — allowing someone to find and potentially schedule an appointment without first navigating through an “appointments” page.

If you hear this and think, “that sounds great for the user, but complex for our editorial team,” you are absolutely and completely correct. The balance of providing ease to your visitors is that someone has to maintain that technology. It’s a new system to configure, and it’s a new domain to govern. And while we’ve worked with tools like ai12z to understand the actual editorial lift (which, in most cases, can be implemented in phases to allow for growth within your team) the setup makes it so the best use case for an AI assistant is when interactions can scale. If you are booking two or three appointments a week, this might be overkill, but if you are a children’s hospital booking hundreds of appointments a day, this is exactly the right way to streamline for everyone.

Regardless of search solution: content is key.

You can choose to revamp your site search, or you can choose to implement an AI search assistant — but regardless of your choice, one thing is key: your content.

Search can't compensate for fundamentally bad content — after all, your content is what’s being searched — so any search project begins with a deep look at the pages on your site. If you have thin, duplicated, or unstructured content, any search solution will reflect those issues back to your visitors.

Here’s what rough content looks like when it surfaces in search results:

  • Two or three nearly identical pages show up at once, with no way to tell which one is current
  • A result technically matches the search but hasn't been touched in years, so the information is incorrect
  • A page is titled in your internal language — "member services portal" — but the visitor searched for "pay my bill"

No tool on its own can fix this, so before you choose between better search and an AI assistant, make sure you are answering the right question in the first place — is this a content problem or a search problem?

Answering this question can save you a lot of time, because the work that fixes your content will often fix your search — and, even if you make a new technology decision based on your new content, it will work significantly better, providing a stronger return on your technology investment.

Structure your content. Clean up duplications and redundancy. Write pages that answer your visitors’ questions. Give those search tools something to work with, and you’ll find search working a lot harder.

How do we make our site search decision?

If there’s one thing to take from this article, let it be to follow your visitors. Don’t start by comparing products — instead, watch what your visitors do. Pull site search logs and look at the terms they’re searching for. They tend to fall into two piles:

  • People trying to get somewhere.
  • People trying to get an answer.

That will tell you what your mix of search queries looks like. Then, look hard at whether your content is written, designed, and structured to provide usable answers or wayfinding. If your content is ready to answer questions, then an AI assistant might be the right choice. If it’s a bit of a mess, your project begins with some serious work on your actual site content.

Selecting a site search option isn’t easy. But, it can be made a lot easier when you look to the source. Think about the people, and then think about what those people want. Do that first, and the choice between better search or an AI assistant becomes the easiest part of the process.

A stylized background with the a i 1 2 z logo.

ai12z: our AI search tooling partner.

We're thrilled to be an ai12z partner — furthering our commitment to solving complex web problems. Need a better way for users to find content on your site? ai12z is the solution you're looking for.

Learn More about ai12z.

Related articles.

AI's a hot topic, and only getting hotter. Check out more articles about automation, generative search, and the modern AI revolution.

Strategy Icon - Castle Chess Piece

Web Strategy

Content strategy, information architecture, governance, and CMS planning — the strategic decisions that make complex web projects succeed.

Platform Icon - Circle with Laptop, Tablet, and Phone

Digital Platform Selection

Choose a CMS based on how your team works — not feature lists. Platform-experienced, vendor-neutral guidance for complex organizations.

Optimization Icon - Speedometer

Digital Optimization

Get more value from your website through strategic planning, performance auditing, and iterative improvement.