When implemented in a smart way, AI search can deliver outstanding results
AI search sounds great in theory, but it has flaws. We know firsthand, since it's one of the many AI integrations we've worked on with our publishers.
Thanks to natural language processing (NLP), AI is very smart about detecting patterns and establishing links between words. The more data it's trained on, the better it gets. For media companies and brands, training AI on your own content can lead to a powerful engine that is ready to search. There's no better example of that than what we did with Raw Story, creating an editorial chatbot based on 300,000+ pieces of content.
But there are some areas where AI search leaves a lot to be desired. Maybe over time these things will be smoothed out, but for now you need to know about them to deliver the best user experience.
Let's break down the pros and cons further.
✅ Pros
- Speed: AI excels at delivering results incredibly fast.
- Contextual Search: Thanks to its language capabilities, you can ask AI questions in natural, conversational language, and it will provide responses that are more likely to address your query.
- Flexibility: AI is highly skilled at understanding context. If a query contains a typo, AI can still find the correct results. Similarly, if a user asks a very specific question, AI may broaden the scope slightly to provide related and relevant content, enhancing the overall search experience.
❌ Cons
- Exact Matching: AI struggles with exact matching. Instead of delivering results that match specific keywords exactly, it often provides varied results. For users accustomed to traditional keyword-based search, this can create a frustrating user experience.
- Outdated Information: Unless explicitly instructed or filtered, AI may surface outdated information, especially if the prompt does not specify a time frame or the content lacks programmed filters to prioritize more recent data.
The Best Approach
So what's our proposed solution? Use AI to do what it's best at, and then use traditional search for the rest.
Our recommendation has evolved. We were once all in on AI search as being a superior search experience. But the data and user feedback is clear — it's not always quite right.
There's nothing wrong with a more traditional search experience, but layer in the following and it becomes really powerful:
- Use AI to check for typos and correct them.
- Use AI to find similar keyword matches and capture intent, instead of using it purely for finding content.
- Use AI to check whether a keyword is intended for semantic or lexical search, and choose the best tool to use.
When it comes to single-word queries, we also made a change there. Use AI for multiword queries, but regular search for a single-word query, and the results are much improved.
Another common complaint is that you can't search by author name in AI search results. This is another area you can make adjustments. Either feed that information into the AI and train it on authors, or add a widget to show authors that match search results.
At the end of the day, yes, use AI. Don't be afraid of its shortcomings. But also be aware of them, and make adjustments so that your users have the best experience possible.
