| | Friday reading: OpenAI launches SearchGPT
OpenAI has announced its long-awaited entry into the search engine race. Rumours had been swirling for months about the possibility, particularly as the company was reported, per The Verge, as "aggressively trying to poach Google employees for a team that is working hard to ship the product soon".
The rumours reached a peak ahead of Google's I/O developer conference in May, prompting OpenAI's Sam Altman to deny that OpenAI was about to launch a search engine on the eve of the event. (The launch announcement turned out to be GPT-4o). However, this time, the rumours are true. OpenAI has stated that SearchGPT is a "temporary prototype of new AI search features", the best of which will eventually be incorporated into ChatGPT.
The prototype will be accessible to an initial 10,000 users at launch, with a waiting list currently open to sign-ups (however, you need to be a ChatGPT user in order to join the waiting list). SearchGPT's demo videos look like standard fare for a generative AI-enhanced search engine, demonstrating that the engine can respond to more complex, multi-part questions such as "When can I see nudibranchs in Half Moon Bay this weekend?", complete with source links. It can also answer directly linked follow-ups such as "will it be hot?" without needing the context reiterated.
What OpenAI is likely hoping will set its new search tool apart is its relationships with publishers: the announcement states that OpenAI is "committed to a thriving ecosystem of publishers and creators", and that it "hope[s] to help users discover publisher sites and experiences, while bringing more choice to search". For the past year, OpenAI has been working to ink details with publishing entities like Axel Springer, the Associated Press, and Dotdash Meredith that enable the tech company to train its AI models on output from these publishers, while also showcasing and citing these publishers more prominently in ChatGPT's responses.
OpenAI's SearchGPT announcement features a quote from The Atlantic's CEO, Nicholas Thompson, who states that, "AI search is going to become one of the key ways that people navigate the internet, and it's crucial, in these early days, that the technology is built in a way that values, respects, and protects journalism and publishers."
In the wake of the news about Reddit blocking Bing (see story below), an array of existing partnerships could prove something of a trump card for OpenAI. But while speculation is swirling about a new threat to Google, will this make much of a dent in Google's search dominance? A recent study from SparkToro in association with Datos indicated that so far, AI-enhanced search competitors have not meaningfully dented Google's dominance . Google's own missteps with generative AI summaries don't appear to have harmed it either, probably due in large part to how quickly Google rowed back the problematic results. Of course, OpenAI is also working with a technology prone to hallucinations and issues. A lot will depend on execution when it comes to how seriously this new player can threaten Google.
Slop begets slop (are AI models eating their own tails?)
If you’re a Marketing Week subscriber, you may have read Econsultancy’s Rebecca Sentance this week summarising the impact of generative AI on SEO.
Sentance explains that generative AI tools have given rise to new opportunities for gaming search engines with spam. She writes that, “While spam and content farms have been a reality of search for about as long as SEO has existed, the magnitude of the problem in the age of generative AI poses unprecedented challenges for Google.”
The dangers that content farming poses to the AI models themselves were explored in a Nature paper published this week, titled, ‘AI models collapse when trained on recursively generated data ’ (Shumailov et al.). As the abstract explains, the authors “consider what may happen to GPT-{n} once LLMs contribute much of the text found online.” It goes on, “We find that indiscriminate use of model-generated content in training causes irreversible defects in the resulting models, in which tails of the original content distribution disappear. We refer to this effect as ‘model collapse’…”.
Simply put, the more AI-generated content out there, the less likely a model being trained on it can find the nuanced information it needs. Think of a snake eating its own tail, or in technical terms, a recursive loop. See The Register for an excellent writeup.
Google’s big U-turn on cookies
The talk of the marketing world this past week has been Google's sudden announcement that it will not be deprecating third-party (3P) cookies after all, just a few months after the latest delay to the deprecation timeline, which had pushed the date for phasing out cookies back to early 2025.
Google was unquestionably facing some issues with the sunset of 3P cookies in Chrome, with regulators and industry partners alike raising concerns about its proposed Privacy Sandbox replacement. In its cookie 'U-turn' announcement, Google revealed that it would keep Privacy Sandbox going, meaning that it will be available as an alternative to 3P cookies, but won't be the only option for marketers to target Chrome users.
This may be more acceptable to the competition authorities, such as the UK's CMA, who had flagged issues with Privacy Sandbox - but may not please privacy regulators like the ICO, which has already said in a statement that it is "disappointed" with Google's change of plans. AdExchanger has a nice piece on what the cookies U-turn could mean for Google's relationship with regulators.
But what about advertisers and marketers? Many marketers will no doubt be breathing a huge sigh of relief as the looming pressure of the (admittedly often-postponed) cookies deadline is lifted. However, a lot of energy and investment has been put into making alternative plans, and while many of these preparations - such as to shore up reserves of first-party data - are still valuable, they were designed with a different marketing landscape in mind. Speaking to Internet Retailer about what the change will mean for retail media , Colin Lewis said that "Many retailers might have believed that their first-party data would be hugely valuable in a cookie free world. Now those views will have to be revisited."
However, most marketers still agree that the 3P cookie is 'crumbling', not least because Google's announcement also included the detail that it will be launching an "experience" in Chrome that gives users an "informed choice" about whether to enable or disable cookies, thereby "elevat[ing] user choice" in Google's words. It's as yet unknown what this will look like in practice. It could be similar to the advent of Apple's ATT (App Tracking Transparency), which had a huge impact on signal from iOS apps. Or it could be more akin to the existing cookie notices on websites that most users ignore. What form this prompt takes could determine how close we come to a world in which cookies are all but deprecated in Chrome - which may mean that relying on continued signal from cookies is still a risky bet. For now, we'll have to wait and see.
Further reading: Google’s U-turn on third-party cookies: What next for marketers and advertisers?
Reddit blocks Bing
Reddit has confirmed that it blocked crawlers from Bing - and many other search engines and AI tools - through an update to its robots.txt file. This is part of the site's ongoing push to ensure that its content is only featured in generative AI if the company behind that AI has paid to license it; since Bing has Microsoft Copilot, which uses generative AI to summarise search results, it has been included in the block. Google's crawlers have not been blocked, which some have speculated is due to the company's existing deal with Reddit.
In a statement to The Verge, however, Reddit denied that the move was "related to our recent partnership with Google" but said that "We have been in discussions with multiple search engines. We have been unable to reach agreements with all of them, since some are unable or unwilling to make enforceable promises regarding their use of Reddit content, including their use for AI."
One way or another, Reddit has obviously come to an arrangement with Google. Microsoft has responded that it "respects the robots.txt standard" and will "honor the directions provided by websites that do not want content on their pages to be used with our generative AI models". The move to block Bing's crawlers has also impacted smaller search engines that use Bing's search technology, such as DuckDuckGo.
It's difficult to blame Reddit for sticking to its guns on AI content licensing, but given that generative AI is now increasingly embedded into search engines, this marks a shift in the way that search may function in the 'age' of generative AI. If more sites and publishers decide that the need to pay to license their content extends to search engines using generative AI, we may see more of a distinction between which sites different search engines can crawl - one that may continue to favour the biggest entity with the deepest pockets. |
Fast Track to Digital Marketing: Next intake, September 17th
Fast Track to Digital Marketing is an 8-week training course covering the key digital marketing topics every marketer should understand. Your team will come away armed with the latest techniques to drive growth and be able to: understand key issues, frameworks and strategies; communicate confidently with stakeholders; and benefit from practical skills that can be immediately actioned . |
|
|
|
|
|
| The Marketing Capability Leaders Forum
September 26th, Waterloo, London, 9am GMT
At Econsultancy HQ, we’ll be hosting our Q3 meet-up for senior leaders in marketing and ecommerce. You’ll hear from a panel of marketing leaders as they reveal how they have navigated transformation, placing capability at the heart of their strategies.
Please note, the forum is invite-only, so registrations are subject to approval. |
|
|
|
|
Econsultancy London
Floor 14, 10 York road,
London, SE1 7ND, UK
+44 20 7970 4322 | | Econsultancy New York
205 Hudson Street, 7th Floor
New York 10013, United States
929 577 9564 |
|
|
|
|
| Econsultancy Pulse is a round up of the latest action in the world of marketing and ecommerce. |
You are receiving this email to [email protected] as you are opted in to receive the Econsultancy Pulse. You can change your email settings or unsubscribe from the Econsultancy Pulse.
Econsultancy is part of Xeim Limited, part of the Centaur Media plc group
Registered in England and Wales with number 5243851
Registered office at Floor 14, 10 York road, London, SE1 7ND
Copyright © 2024 Centaur Media plc and / or its subsidiaries and licensors. All rights reserved. |
|
|
|
| | |
|
