AI site search, AI chatbot, or traditional search: a decision framework

If you work in digital, marketing, or IT at a university or council, you've probably had this conversation in the last six months. Someone forwards an article about an AI search tool. Someone else has been impressed by a chatbot demo. A third person points out that the existing site search is, by all measurable accounts, fine. And the question lands on your desk: which of these do we actually need?

The honest answer is that they do different jobs, and most large content estates need more than one of them. The harder answer is working out which.

This article is a framework for that decision. There are use cases where AI site search is genuinely the wrong choice, and we'll be specific about those. The goal is that you finish reading and explain the difference to a head of service, a procurement lead, or a senior stakeholder, and which is best without reaching for a supplier pitch deck.

• The decision framework
• Traditional site search
• AI Chatbots
• AI Site Search
• Comparison table

For most of the web's history, three jobs were clearly separated. If you wanted to find a known document, you used search. If you wanted to read about a topic, you used navigation. If you wanted help with a task, you called someone or filled in a form.

Generative AI has blurred the boundary between answering a question, holding a conversation, and looking something up. A modern AI tool can do all three, sometimes in the same exchange, which is why the category labels have stopped being useful. "AI search" can mean a retrieval-augmented question-answering system, a conversational agent, or a souped-up keyword index, depending on who's selling it.

So before deciding what to buy, it helps to separate the jobs.

Six questions, asked in order. The answers route you to one of three tools, or to a combination.

If they know the title, document name, or reference number of the thing they want, traditional search wins. If they have a question in their head but no idea where the answer lives, AI site search is the better fit. If they have a task to complete and want guiding through it, that's chatbot territory.

Structured, well-tagged, list-like content (course catalogues, councillor directories, planning applications) suits traditional search. AI site search suits content that varies in length and structure across the site — clean markup still helps, but it doesn't depend on the uniformity traditional search needs. Chatbots suit interactive workflows where the answer depends on inputs gathered through the conversation, like eligibility checks or benefit calculators.

A ranked list of links? Traditional search. A synthesised answer with citations? AI site search. A back-and-forth that ends in a completed action? Chatbot.

AI site search needs you to care about what your content says, because the system will surface it as answers. Chatbots need scripted intents, fallbacks, and review cycles. Traditional search needs the least ongoing oversight, which is one reason it has lasted.

Below roughly 200 pages, AI site search is overkill and traditional search is usually fine. Between 200 and a few thousand pages, AI search starts to earn its place. At university or county council scale (tens of thousands of pages across faculties, services, and microsites), AI search is often the only thing that can navigate the sprawl.

If the user needs to report a missed bin, pay a bill, apply for a course, or book an appointment, none of these tools are the right primary answer. A well-designed form or a focused chatbot beats search every time for transactions. Search gets them to the form; it shouldn't try to be the form.

Traditional keyword search has been written off prematurely. For known-item retrieval, where the user knows what they're looking for and just needs to get to it, a good keyword index with decent synonym handling and faceted filtering is hard to beat. It's fast, predictable, cheap to run, and the user understands what just happened.

It still wins for:

• Structured listings: courses, councillors, news articles, planning applications, FOI responses
• Expert users who know the terminology and want to filter results themselves
• Content where exact wording matters, like statutory notices, policy text, or legal documents
• Sites where most arrivals already know the document title or reference

The failure mode of traditional search is well documented. Users who don't know the right keywords, or who phrase their question conversationally, get nothing useful back. That's a real problem at scale, but it's not a reason to replace traditional search everywhere. It's a reason to add something that handles the other cases.

Chatbots are for getting something done. Reporting an issue, booking a slot, completing an application, navigating a decision tree to the right team. The good ones are scoped tightly to a handful of transactional flows and tested rigorously against the intents they support.

What chatbots are not good at, despite frequent claims to the contrary, is answering open-ended questions across a large content estate. They can be wired up to do it. But the result is usually worse than either AI site search or a well-organised navigation. The conversational frame implies a depth of understanding the system doesn't actually have.

AI site search sits between the other two. It accepts natural language questions, retrieves relevant content from across the site, and synthesises an answer with citations back to the source pages. The user gets a direct response to what they asked, not a list of ten links they have to triage.

It earns its place when:

• The content estate is large enough that browsing isn't realistic
• Users arrive with questions rather than keywords
• Queries have spelling mistakes or are written in another language
• The content already exists and is reasonably accurate (AI search will surface what's there, including the bits you'd rather it didn't)
• You can sustain the governance burden of caring about answers, not just pages

Four cases where we'd actively recommend against it:

1. Small content estates. Below a couple of hundred pages, the user can find things by browsing, and the cost of implementing AI search isn't justified.

2. Transactional journeys. If the goal is completing a form, paying a bill, or booking a service, a focused chatbot or a well-designed form will outperform AI search. Search is a way in, not the destination.

3. Regulated or legal content where exact wording matters. A synthesised answer is the wrong format for statutory notices, planning conditions, or anything where the user needs the precise text and nothing else. Link them to the source.

4. Sites where users overwhelmingly know the document title. If your analytics show people typing course codes, councillor names, or document references, traditional search with good filtering will serve them better and faster.

Being clear about these cases is, frankly, more useful to a stakeholder conversation than another list of features.

Almost every large estate we see ends up using two of the three, and a few use all three. A university typically needs traditional search for course listings and staff directories, AI site search for prospective student questions and research content, and possibly a chatbot for clearing or applications. A council typically needs traditional search for planning applications and councillor lookups, AI site search for service questions, and chatbots for the highest-volume transactions like missed bin reports.

The mistake is treating this as a single procurement question. It's three jobs, and the right answer is usually to be deliberate about which tool does which.

Whichever combination you land on, the question your stakeholders will actually ask is whether the thing is trustworthy. For AI site search specifically, that means citations back to source content, the ability to see what was retrieved and why, and confidence that the system won't invent answers when it doesn't have the content to support them.

Ryan Bromley's piece on implementing AI-powered website search without losing trust covers this in more depth, and Richard Chivers' analysis of why traditional website search is no longer fit for purpose is a useful counterpoint if you're trying to build the case internally.

If you've read this and concluded that AI site search is part of the answer for your web estate, the Insytful AI Search product page has the details on how it works. If you'd rather start with the framework, the one-page PDF below condenses the six questions into a flowchart you can take into a stakeholder meeting.

The honest test of an article like this is whether you can now explain the difference to a colleague without using the word "AI" more than once. If you can, it's done its job.