How SearchGPT Might Finally Dethrone Google and Redefine Digital Marketing. And How You Can Prepare. 

By Adam Hanft and Ed Sussman

If Google is earth, then SearchGPT is an alien invasion, and earth is in for a major war. You may be a prosperous denizen of the big blue marble, having spent decades learning the intricacies of and peculiarities of how to be successful in earth’s relatively stable environment. Get ready to duck and dodge because AI cyborgs may destroy your branding, your reputation and your website traffic if you don’t learn to understand them, make friends with them, and adjust to their entirely different worldview.

Is Google’s dominance threatened?

Google, of course, knocked out every meaningful search competitor more than 20 years ago. It’s been so dominant that on August 5, 2024, a federal judge ruled Google search was monopolistic.

The ruling will give OpenAI even more momentum following its July 25, 2024 rollout of SearchGPT. SearchGPT is different from previous Google competitors. It’s not just the $10 billion in Microsoft dollars behind it. SearchGPT may actually deliver on the original promise of search: a savvy answer machine, not a list of links leading to websites of wildly varying quality. Great for users but potentially disastrous for marketers dependent on Google links for most of their traffic.

And when AIs replace links with answers, they become publishers. As with any publisher, if they get a story wrong, or don’t mention your brand in a round up, your reputation is going to suffer.

At least 4 massive AI armies (and lots of promising smaller ones) are attacking Google search: ChatGPT, Meta.ai (which fully integrated AI search and chat into Instagram, Facebook and WhatsApp) and Microsoft CoPilot, soon to be joined by Apple’s iPhone with next-gen Siri search powered by ChatGPT. Money is pouring in. Google’s $175 billion search business will not come up for grabs again.

Google has been preparing for the AI search war for at least two years, ever since the media frenzy surrounding the November 20, 2022 release of the flawed but exceptionally capable ChatGPT 3.5.  For a full year, Google cautiously tested Google Search Generative Experience (SGE) in a public beta Feeling confident, on May 20, 2024, Google released Google AI Overview, the rechristened SGE, aggressively positioning it at the very top of almost every traditional Google search result in the United States.

The world was not delighted. When people go to Google, they still want the familiarity of terra firma. Nothing on Google.com homepage is going to be treated by its users as an experiment, even if it’s labeled as such.   Social media and the press quickly spread examples of severe Google AI misfires, e.g. advising a user to eat one small rock a day for nutrition, and suggesting to a user that the Golden Gate bridge was the best choice for a suicidal leap.

Google retreated fast rather than further risk their reputation. AI Overview might appear on roughly 5% of searches in the U.S.; more frequently if you phrase your search as a question. But AI Overview is not like the ill-fated “Google +” social network. Walking away from AI search would relegate Google search to being a landline in a world of mobile phones. It cannot back off and is continuing to aggressively expand AI Overview globally.

Despite Google’s search dominance, ChatGPT (which will soon be integrated with the best features of SearchGPT) is not that far behind. Back in 2023, ChatGPT was the fastest growing application in the history of the internet (including the internet itself), and that was with plenty of error-laden answers and without integration of the live web.  Currently it has 180 million active users,   That’s without free live web integration, which promises to be a killer feature. Google search has 2.45 billion users, a not insurmountable lead.

ChatGPT’s usefulness will predictably trend up every year for the foreseeable future, whereas traditional Google link search is already at its technical zenith.

Google’s multi-purpose free GenAI platform, Gemini, is meant to compete with ChatGPT and limit its growth. It had 271 million visits in June 2024, down from 376 million in May, according to SimilarWeb (probably fewer than 60 million unique visitors), but it’s been integrated with the live web from the start. So it’s not about to get a big boost when it integrates the live web, like ChatGPT.

And that antitrust case Google lost? Depending on the punishment phase outcome, it might force Google to drop their search deals with browsers and phones other than their own. 

How does AI search work?

Eventually, AI search will work by making search obsolete. In the next year or two, with just one short request, an AI personal assistant will be able to recommend your next vacation destination, book your hotel and airline tickets, set your itinerary, make restaurant reservations, and order your favorite sunscreen before you leave.  It’s also going to be the tour guide in your ear and instant translator when you arrive, perhaps in the soothing voice of Judi Dench.   If you have a question about what concerts are going on and what time a show starts, just ask (out loud if you prefer) and you’ll get your answer. No clicking around.

In other words, search will no longer just take place on search engines. Searchless search. In most cases, it will be baked into virtually all AI applications, making traditional search links a niche product for in-depth research and people who don’t trust AI. Google’s two decades of brand loyalty and ingrained user behavior have bought it both time and a legacy user base that will be reluctant to move on.

SearchGPT threatens Google precisely because eventually its best features will just be ChatGPT, with its myriad of other applications, and steady increase in utility.  It’s debatable where OpenAI will see the need for a separate SearchGPT product after ChatGPT is up to speed and it takes its next technological leap.

The revolution in searchless AI search has also moved somewhat slowly the past two years because highly complex accuracy and reputational issues still need more work. And because AI hasn’t quite moved for long enough along yet as a personal assistant. It shouldn’t be long. OpenAI’s “Project Strawberry”, expected soon according to social media leaks, may have the next level of predictive analysis necessary for turning AIs into assistants that anticipate your needs and can act for you as an “agent” without needing detailed instructions.

As for accuracy, we know from the paid version of ChatGPT4o, which is already integrated with the live web, that OpenAI has made substantial progress in returning relevant, accurate results. It turns out the need to support a statement with source attribution and links also reduces errors. Yes, it’s still prone to bias, misses critical information, and repeats errors from poor sources, but its users don’t expect it to be Google. A certain level of problematic responses in exchange for the coolest technology in a generation is just part of the expectation for a challenger brand like ChatGPT.   Google AI Overview is going to need to get its accuracy to an exceedingly high-level to be embraced by the public like Google search over the past two decades.

What about right now? How is SearchGPT different from Google.com and Google AI Overview? 

To start, the page is a lot cleaner. One query or voice command on SearchGPT, and you get one answer. No clicks. SearchGPT has reduced choice to increase usability. Yes, there are responses with attributions and links, but only where it makes sense. If your question has a clear answer, you get the answer with one or two highly relevant links.

On Google AI Overview, you might get a concise answer, but you need to click to see all of it. Google has been experimenting with how many links to provide, but 20+ links is common somewhere within the answer is common. An array of other search features and links crowd the page. One search, many answers.  And if your search doesn’t trigger a Google AI Overview, then you get regular search results.

Google search is fighting with itself – is it trying to index the web or is it trying to give the user the best answer? After 25 years, the two are not the same thing anymore.  Yet Google can’t just stop being Google all at once. It’s a company built on lots and lots of links, with tens of billions of dollars in link-related revenue to protect. Aside from Google.com revenue, there’s all that revenue from the ad network it runs on millions of websites. If it transitions too quickly, it not only risks its own revenue, but could crater traffic for the entire world wide web. That’s a consequence of running a monopoly.

Critically, like any good AI, SearchGPT actively encourages the user to dig deeper with follow-up questions, making it even less necessary to click on links. The latest version of Google AI Overview has eliminated follow-up chat entirely. Google is effectively saying, if you want more information, you’re going to have to click on one of my links. No more easy answers for you.

That’s good news for websites and publishers trying to keep their traffic up. But it’s a tremendous long-term vulnerability if Google AI Overview doesn’t revert back to the follow-up chat it allowed in beta.  One of the things that makes GenAI great is that you can converse with it.  Yes, Gemini is fully chat enabled, but it has about 2.4 billion fewer users than Google search.  Google’s other attackers such as Meta.ai, Siri and even Bing CoPilot, are not very concerned with providing loads of links. Their AIs will keep retrieving information and carrying out tasks for you

Even if Google.com figures the proper blend between AI responses, links and all its other search features, it is still just going to be search. It might be the sharpest search blade of all, but to have the biggest information retrieval role in the daily lives of the largest number of people, a platform will need to do more than search. Gemini is going to be facing off with ChatGPT for the full AI assistant market, but the chances that Gemini will overcome and dominate ChatGPT and all the other competitors seems small.

If you’re in the brand marketing, search optimization or reputation fields, your instinct might just be to keep you head down until the situation is clearer. But we can tell you from experience the savvy players are getting prepared now.  And they’ve already developed techniques for AI Optimization that work.

How to prepare for the AI upheaval in digital search marketing and online reputation. 

Check to see if you’re already in a good or bad place.

Turn your most important keywords into short questions a user might ask a chatbot.  If you come from a profession like PR or reputation management, instead of keywords, you have to anticipate AI queries that will affect your client’s reputation. Then ask the queries at Google.com (to see if Google AI Overview) is triggered yet;  Google Gemini, Meta.ai (you’ll get the same response there as Instagram, Facebook or WhatsApp), the free ChatGPT4o (even though it’s not live web integrated yet), the paid ChatGPT4o  (which is the live web integrated), and Bing CoPilot.

Unfortunately, you’re going to need help even for an informal check. The GenAIs are going to cache the response they give you, so you won’t get to see what other people are seeing – which is going to be completely different.  Generative AIs are probabilistic. It’s part of the generative AI model that they have tremendous variability and the exact mechanisms of their decisions in any specific case can’t be known. That’s why you’ll sometimes hear them referred to as a black box. If a platform is giving the same answer every time, it’s not using generative AI.

You should work with other people on your team with different browsers and different accounts and aggregate the responses to take a look at the variability.  It’s not an accurate sample, but you might get a right sense as to what’s going on.

Does your brand show up every time? Or never? Are good or bad things being said? What’s missing? Anything biased or inaccurate?

Let’s say you’re a luxury hotel resort in Hawaii, and there are 10 major competitors and 5 challengers. If the AI gives a list of hotels, it’s not going to be in the same order every time, it’s not going to choose the same hotels every time, and it might leave your hotel out every time if it’s on page one of search results. But it’s not random either. Citate research shows that AI consistently chooses some brands to be more prominent, but it might have a wider variance of choices for the next tier.

Early research showed that on Google’s AI beta, AI responses only included page one search engine links page about half the time. The others were challenger brands that might not even be in the first few search result pages. This is a good moment for challenger brands. Every big AI also has its own crawlbot retrieving information separately from search engines. They are not looking for the same thing and they are not using the same methods as traditional Google search when generating a response.

As we said, your spot check is not going to be reliable, so don’t plan an optimization campaign based on a handful of results.  To get a statistically-reliable sample you need a service like Citate which repeats prompts, and close variations, hundreds of times until our own AI shows the data in the answers are converging on an acceptable threshold. This is a difficult enough technical problem that we needed a Harvard AI professor and some very brilliant AI engineers to create  (and apply for a patent) on a reliable methodology.

 But even with some spot checking you might identify either welcome news or a problem. Just don’t neglect to repeat the prompt many times from different browsers and accounts.

Experiment Now with AI Optimization. 

New tactics for AI Optimization are already emerging. Just be careful. Some will prove to be simplistic nonsense and not work, or backfire. There are a bunch of search engine tools already on the market that use AI to create content they recommend you put on your website for search engine optimization. It’s very unlikely they will work for search engines or AI

AIs don’t look for keyword placement or density like old search engines. AIs are smarter than that. They are looking for topics, which might be represented by any number of semantically-related terms. The AIs are good at figuring out when different content has similar meaning. If it looks like your content is repeating  keywords or repetitious even with variants, the Ai is likely to sniff out another AI being asked to write to old optimization techniques. This content might get severely penalized.

The problem with AI-generated SEO content is that it isn’t terribly useful to people. It’s written for robots. The AI companies are devoting tremendous resources to try to separate out the spam from the good stuff. Without this ability, AI responses won’t be better than Russian-bots writing social media posts. Spam and propaganda being regurgitated on AI is already a severe problem and it threatens the credibility of the entire AI ecosystem.  You might just be using a search optimization tool using AI to create copy about a coffee maker, but it might be treated just like disinformation if it isn’t done right.

If flooding your website with AI-generated content won’t work, what will? Some ideas:

• Focus on the topics the AI cares about. Traditional search engine algorithms looked to see if your content matched user intent. Think about AI intent instead.
• Clearly identify the authority of the person writing the content. AIs are going to be able to do a better job confirming that a person is authoritative.
• Live web-integrated AIs are very responsive. If your idea fails to work, try again soon.

It’s early days in AI Optimization. Will social media content help or hurt?  Which publications has SearchGPT hand-selected to favor? (Early: check its testimonials page.)   Our own tactics can work for months, only to face a sudden reversal not characteristic of search. So we are constantly experimenting and updating and releasing new tools. You’re going to need analytics to measure whether your tactics are working. Don’t trust a handful of responses, whether good or bad.

You may need different optimization solutions for different AIs

Some AIs won’t respond to AI Optimization for a year or more because they aren’t integrated with the live web. They rely entirely on a training data set that might only get updates once a year. An example is the current free version of ChatGPT 4o.

Might be the same content needs different sections for different AIs; might need different pages blocked from visibility by all but the one AI you are targeting. Experiment, experiment, experiment. Winners now will have an edge that could mean the difference between thriving and bankruptcy

Monitor for inaccuracy, bias and missing information. Don’t forget about reputation either.

Errors, misinformation, missing information, bias and subjective judgments are commonplace on all forms of generative AI, but especially those that are web integrated.

GIGO.

AIs models which are not web-integrated can’t be optimized with new content until the platform updates. That might only be annually. But even gated AIs can be influenced when it comes to mistakes and bias,

Behave ethically. And report misinformation. 

We already know that the same black hats who pollute social media, Wikipedia and create fraudulent media sites are working on manipulating GenerativeAI.  This is going to be an ongoing, difficult problem for all the AI platforms. They know it could ruin their business if misinformation doesn’t come under better control.

Being positive about your brand or your political agenda is not misinformation. We have not seen AIs penalizing companies just for being enthusiastic. However, when one website represents facts in a manner directly contrary to many other credible websites, we already know from our statistical modeling that the underlying model is likely to ignore the outlier fact. No person needs to make this decision – the Large Language Model trains itself to measure probability this way. It’s possible that pressure from regulators, the press and social media, as well as competition  between platforms, will cause an overcorrection in refining models to penalize websites publishing factual outliers. This is something to monitor carefully if there are disagreements between you and other sources.

Correcting misinformation on your website should not be penalized. But you have to be very careful how you do it. Citate recently picked up Google Gemini spreading horrendous misinformation derived from the client’s own methodically-documented FAQ page on their website. A Wikipedia editor then repeated the misinformation. AI is not an acceptable source for a Wikipedia article, but rumors can be repeated in discussions between editors behind articles. And some AIs train on all parts of Wikipedia, leading to a reinforcing mechanism for AI misinformation.

The client produced the website page for the press in order to counter misinformation being spread on social media.  While there are mechanisms to report misinformation on all the AI platforms, in this case, the client had full control over the problematic page. The solution was to rewrite the website page so each myth was clearly identified as such instead of repeated, then debunked.  Perhaps by the end of the year, most AIs will be able to analyze content carefully enough to determine context in a situation like this. As of now, it pays to monitor and correct problems now,, before they get reinforced.

Regular reporting of misinformation does help.The big AI platforms are all using machine learning and “humans in the loop” to improve their models.  Including a link to a reliable source that confirms the AI is posting misinformation should help.

Break down marketing and reputation silos. 

The previous example of AI repeating severe misinformation from a client’s own attempt to clear it up for the press is a good case study of why silos between marketing departments are going to need to be broken down and re-arranged.

AI is a publisher of stories, not just links.  Search engineers are accustomed to competing for link rank, not shaping narratives.  Optimizing AI for ROI is not going to include reputational monitoring or remediation for reputational issues. And PR people are not going to be accustomed to setting up and monitoring analytics campaigns that have more in common with search engines than counting audience reach.

Going forward, an organization’s team members will need to know that reputational issues being addressed via social media, website content or the press also need to be monitored across AI. And content will need to be phrased in a way that evolves to reflect that ability of AI to pick up on tone and context.

Conclusion

Whatever happens in the war between Google, SearchGPT and the other competitors is going to dramatically impact you, too.  While it’s useful to think about long-term societal and business shifts that will be caused by AI, when it comes to digital marketing and reputation, any organization even thinking about about early 2025 should begin to prepare now.

As for accuracy, ChatGPT started out in terrible shape when version 3.5 was released in November 2023. But now we know from the paid version of ChatGPT4o, which is already integrated with the live web, that OpenAI has made substantial progress in returning relevant, accurate results. It turns out the need to support a statement with a source attribution and links does reduce errors.

With so many fewer links of SearchGPT (and eventually GPT), there are winners and lots of losers.  The same is true Meta.ai, another pure-play multi-purpose AI platform that does not need to worry about disrupting its current business model. And it’s going to be true of Siri almost by default. A talking AI may show some associated links, but these will be far less important than what it’s saying.

By contrast, even if Google figures out the proper blend between Google AI Overview and Google search, it’s still just search. It might be a very sharp blade, but it’s just that. Will it help you with your homework? Or remind you to buy milk or engage in a philosophical debate with you? No. It’s just been specialized for search. To get a multi-functional AI from Google you have to head over to the side stage of Google Gemini. Gemini is not that great at search now, though it’s ok at finding select important links, But everyone knows the real show at the main stage is Google.com. Google needs to save Google with Google.com.