Key Takeaways
- Google’s integration of generative AI in Search faces issues with incorrect answers.
- Google risks reputational damage with AI Overviews, as incorrect results can harm user trust.
- ChatGPT search and Microsoft’s competition may impact Google’s AI strategy.
It’s been a while now since Google announced its biggest change to Search in years: making use of its new generative AI tools to provide an overview of the topic being searched, so that we no longer need to click into a resulting link to find out the information we need. Sort of. The integration of new AI features hasn’t exactly gone down well, immediately causing all kinds of problems with incorrect answers, ranging from the absurd to the potentially dangerous.
Google is putting a lot on the line with its changes to Search, and this appears to be indicative of the pressure the company is now feeling from both Microsoft, ChatGPT, and Apple, striving to bring better AI integrations to its products.
What is generative AI search?
Announced at Google I/O in May, AI Overviews is a previously experimental feature of Google Search that makes use of generative AI to answer some search terms and questions. Over the last few weeks, it’s been rolling out to everyone across the United States, with heavily mixed results. Google is reportedly relying significantly on a deal with Reddit that they struck to use its content for AI-generated datasets, and users have reported individual Reddit comments being the sole source for some of the answers generated, as well as a plethora of other incorrect, misleading, or downright silly replies to simple questions.
Google is taking a huge risk
Integrating generative AI with Search increases the risk of reputational damage
Most AI features in core cash-cow products have been ‘nice to haves’ – additions to basic existing services. By placing AI’s response first and foremost in Search results, Google is taking this a bit further, signaling as surely as possible its intentions for AI to take the stage, even at the risk of compromising the reputation and integrity of its most fundamental product.
This seems, at the very least, dangerous. Gemini has failed to take off in the face of ChatGPT, which has arguably been the first serious Search competitor since Bing to break through to the masses. But ChatGPT has problems, as does Gemini, and until now any reputational damage from those shortcomings has been constrained to those products. Now, with an AI response integrated directly into Google Search, which is the default search tool for just about every browser on earth, it gives a lot more eyes, ears, attention, and opportunities for nasty hallucinations.
Apple has taken a similar approach, integrating both its own models and ChatGPT into iOS with Apple Intelligence, but these are firmly nice-to-have additions to an existing OS that are unlikely to drive users away from the platform towards competitors.
Related
Incorrect results are damaging the entire search experience
Search is arguably Google’s most reputationally risky product. Previously, Google Search results would surface links to information, on a basic level, guiding us to sites and summaries of other people’s content. But now, generative AI Search is taking responsibility not only for surfacing information, but also summarizing and potentially editorializing it – and we’ve already witnessed some of the mixed results. In the aftermath of AI Overviews rollout, Google was forced to make changes to improve results, and clarify the root of some of the more egregiously wrong summaries.
There’s also the question of the audience. Google is visited over 80B times a year, and holds over 90% of the total search engine market. While Gemini and ChatGPT have gained some public attention, the sheer volume of queries, edge cases, and potential for damage is exponentially higher. Even iOS AI integrations are likely to be used on a scale several orders of magnitude smaller than Google Search.
The combination of these two factors creates a legitimate risk of reputational damage to Google Search. Incorrect results or hallucinations will be publicized quickly, and beyond the hit to Google Search’s reliability, potentially will damage broader consumer impressions of AI and lead people to question the reliability of AI generation.
Related
ChatGPT search is rumored to be coming
Microsoft is once again Google’s biggest competitor by proxy
The elephant in the room here is the rumored ChatGPT search. As we alluded to earlier, ChatGPT has broken out of techy-spheres and infiltrated everyday life for many of us. It’s in use for everything from programming, to data processing, to writing cover letters. It’s also been rumored to be working on its own search product.
The uptake of ChatGPT, and the ubiquitous news attention any new designated search product from them may get, could risk leaving Google stuck with a standing start when it comes to integrating their products more widely. There might also be the ‘brand new’ effect of ChatGPT to consider – it could be perceived as more cutting edge than Google’s product, which is almost as old as a decent percentage of its users.
Integrating search now might be Google’s attempt to get ahead of the narrative on other AI-powered search alternatives, even if it needs to ride out some teething issues during its roll-out. This is also a good opportunity to familiarize the wider Google user base (as people outside of tech-blogs use Google) with the power of AI Assistants, and offer some tangible and real improvements to the wider search experience (when it works).
Whatever the outcome, Google is acting boldly
It’ll be difficult to judge how effective it will be to have generative AI integrated into Search, especially after its shaky start. But one thing is clear. Google is taking decisive and bold action around AI, not only investing in developing top-tier LLMs, but placing them firmly front and center in its most important product. This is a huge risk, but it might be necessary to head off the threat of ChatGPT – a threat the scale of which Google hasn’t faced since Bing tried and failed to compete in the early 2010s.