ChatGPT: What Now for Nonprofit SEO?

But it wasn’t just a pun: even back then, the talk was of how machine learning was the future disruptor of SEO. I didn’t know that it would take a further seven years for this AI-centred disruption to fully arrive in a way that went way beyond the usual “this is the thing in SEO this year” (see: voice search, mobilegeddon).

With ChatGPT, the talk across all digital industries is that this disruption is very much here. This blog explores some early thoughts on what this means for nonprofit SEO.

(We’re at the bit in the film where Sarah Connor is on her way to Miles Dyson’s house, for those of you who still want the analogy).

You already know what ChatGPT is. The world doesn’t need another explainer, but there are two important things to note.

Firstly, no one envisages ChatGPT in its current form to replace Google’s search engine model. What you see today when you use ChatGPT should not be taken as the “replacement” to how users currently search: these large learning models will be integrated into the search experience in ways that will look vastly different from what’s been released so far. Remember GeoCities? Altavista?

Secondly, while ChatGPT is the buzziest AI chatbot right now, it is not the only one, and it will definitely not be the last. In a recent leader, The Economist noted, “not a week goes by without someone unveiling a “generative” artificial intelligence (AI) based on “foundational” models—the vast and complex algorithms that give ChatGPT… it[s] wits”. During the writing of this blog, OpenAI released GPT-4, their newest AI language model, replacing ChatGPT-3. Microsoft opened up public access to its integrated ChatGPT-4 powered search engine. ​​Google launched an API for PaLM, allowing access for developers to its own AI language models, and updated its Workspace apps (Google Docs, Sheets etc) to integrate AI features. There is a pace here that is without doubt new.

With these two points in mind, there are a number of concepts that lie at the heart of ChatGPT in its current form worth exploring right now for anyone working in SEO at a charity.

One of the most interesting areas of ChatGPT is how it responds to queries where multiple sources of “compounded” information is required for a coherent response.

Imagine a user with £1,000 to donate to charity. Epilepsy is a cause they resonate with, but they’re not sure how best to donate their money. A realistic question they might ask is “What's the most impactful way to donate £1,000 to support epilepsy research”.

Google’s SERPs respond to this query with an organic result for the Donate page for the charity Epilepsy Research, followed by a similar result from Epilepsy Society. After that, a “People Also Ask” knowledge panel is displayed. As it stands, Google’s SERP isn’t really answering the question in a meaningful way. In fact, it feels quite crude all of a sudden. ChatGPT, on the other hand, shows what might be possible in this instance:

It’s not perfect, but the nuance and depth of the answer ChatGPT displays far outweighs any traditional SERP. It is highly unlikely a piece of content exists that will rank at the top of Google that currently does that same thing as what ChatGPT does here. Today’s breed of search engine will begin to integrate AI to also provide this kind of compound answer (Google calls it “synthesized insights”). How it does that, while preserving the concept of attributed sources (covered later in this blog) will be super interesting.

Lots has already been written about the role AI will play in the process of content creation. In summary, models like ChatGPT can already create content that is in many cases nearly indistinguishable from what a human might create. For SEOs that means title tags and meta descriptions, but also full responses to things like content briefs.

What will the cost be here? This feels like an extremely depressing thing to get excited about: the idea that human writing and language can be recreated by an algorithm with nearly no differentiation. Nick Cave has already spoken out on this in the context of art, but this isn’t just limited to traditional creative fields. If AI takes us down a path that results in content being farmed out to automated tools, humans will likely respond to this, and seek out content that engages them again. It’s maybe not so dissimilar to the rise in sales of vinyl albums over the last few years: music lovers now actively pay much, much more for a vinyl that is available on Spotify for free, because they seek tangible difference, and uniqueness. They’re rejecting commodification and placing distinct value on the active process of consuming music, and if the rise in AI-generated content takes off, it’s our bet that humans will instinctively reject that over time.

For charities who are excited about the implications for content creation, we’d urge caution. These tools will most definitely speed up the processes around some routine content creation where tone of voice and style is less important (schema markup is one area that will likely be automated). And it’s already possible for particular models like CoAuthor to use output from existing writers to train itself and create new, stylised content. But this stylisation still needs to be based on real human input. There will be a continued need for charities to invest in people who can write well, because our human responses to good, unique writing will not change.

ChatGPT is trained on a data model that is currently fixed in time. Anything that happened after September 2021 is not included in ChatGPT’s “training.

This concept has major implications, not least around searches that Google terms “Your Money Your Life” (content that impacts “the future happiness, health, financial stability, or safety of users”). Without a method of training itself on more recent datasets, ChatGPT does not provide healthcare information that could be termed “up-to-date”.

This is already changing, and it is already possible to synthesise current, authoritative data with ChatGPT’s ability to understand and generate natural language (Torchbox is working on such a model).

There are some interesting, related time-based distinctions that could, in future, have implications on donor sentiment. ChatGPT is clear on settled historical issues, like whether the holocaust really happened. It does not, however, “take a view” on the Russian invasion of Ukraine.

This is in stark contrast to Google’s results on the same query.

One wonders how an “unbiased” display of information might differently affect donor sentiment for charities who work in humanitarian spaces. The response from the general public to the war in Ukraine was largely fuelled by the lack of complexity around who was right and wrong; the bad guy was clearly defined, in a way that everyone from the BBC to the Daily Mail understood. With ChatGPT, what are the implications here for how the donating public respond to live issues of the day, and (deep, terrified breath) how might this vary from model to model based on the organisations funding them?

One of the funnest things about the lightning speed emergence of ChatGPT has been watching people draw the tool into arguments around the validity of its answers. The point being made is that ChatGPT presents its information back with absolute certainty, because it doesn’t actually understand that information in any way.

Compounding this, ChatGPT doesn’t attribute its information to specific sources, so a user has no way of knowing where information came from.

In today’s search landscape, carrying out a Google search for “does semaglutide make you lose weight” leads the user to a typical set of results that the user then navigates based, in part, on their understanding of the sources that are served up in Google. Lots of users will append a search like this with the word “NHS” or they’ll quickly scan the SERP for a domain result that they trust. ChatGPT provides no attribution for its information, so there’s no way of knowing whether semaglutide really does cause weight loss at “an average of 15-20%”.

It is, to put it bluntly, frightening to read ChatGPT’s responses to even those queries you already have a learnt understanding of, because of the sheer unattributed certainty that the information is presented back with.

As the industry matures and grows around AI-powered search, so too will difficult legal and ethical scenarios. It is easy to imagine a situation where the wrong healthcare information is provided to a user in life-changing ways. In tackling these issues, search engines and policy makers will certainly lean into expertise, authority and trust for source data. E-E-A-T is a core concept of SEO in 2023, so if your charity hasn’t already engaged with it, now is the time to do so, because it’s only going to become more important.

The announcement that Microsoft had invested $10bn into Open AI was one thing.

But the news that Bing (owned by Microsoft) was now running on “a new, next-generation OpenAI large language model that is more powerful than ChatGPT” did one clear thing: it gave Bing a major shot in the arm. Forever the forgotten search engine, those nonprofits in the know understand Bing to be a search engine that delivers stronger results than Google in terms of ROI and CPA. Its weakness thus far has been volume: bluntly, no one really uses Bing.

This could change, and while it’s unlikely this is Google’s “kodak moment”, as some claim, this could shift the public’s perception in game-changing ways, and bring Bing out of the darkness as a respected (and used) competitor to Google. What’s more, it could fundamentally rebalance Bing’s demographic skew towards older users. This is exciting; “how can we reach the next generation of supporters” is the question we’re most frequently asked by new and existing clients. Somewhat unbelievably, Bing might well play a part in the answer if its share of the search market shifts to include younger users drawn to the perceived buzz around its AI capabilities.

At Torchbox, we don’t recommend a specific “Bing SEO Strategy” for nonprofits. But there are things that most charities aren’t doing; one of those is utilising Bing Webmaster Tools, which is the equivalent of Google Search Console. If you’re not already studying and cross-comparing search query data across both SERPs, you’re missing a trick. It’s free, and it contains very specific, quantitative data on how users (with a different demographic profile to those on Google) are looking for your charity.

It is well-documented that Google has been slow off the mark comparatively. This narrative is overstated. Google is not a new player in the AI world; it has not not been looking at this for the last decade, or hoping it all somehow goes away. The folks at DeepMind ( a subsidiary of Alphabet) are not asleep on the job: foundational work will have been done in this space that means although they’ve been slow off the mark, they knew this was coming and they’re probably not far behind.

The question is: how quickly do they now need to move? The “botched” release of Bard, Google’s answer to ChatGPT, suggests some panic internally at Google, but given the (wallet) size of Google, it won’t be long before the playing field is level again. While writing this blog, Google Health held an event in which they discussed their plans to use AI to transform healthcare through its Med-PaLM language model. Google has already announced that Bard will be integrated into search imminently; and invested $300m in Anthropic, a rival to OpenAI, whose ChatGPT equivalent is named “Claude”. Claude is now integrated into Quora and DuckDuckGo.

This race has just begun. It’s unlikely it’ll be another seven years before we talk again: Torchbox is about to release a product which deals with many of these concerns, enabling conversational access to your trusted, up-to-date content.

Find out more about our SEO services here or get in touch via email [email protected]