Welcome to our September Digital Digest!
It’s officially autumn, and whilst the natural world around us enters hibernation the world of digital marketing shows no signs of doing the same.
There’s been headline after headline with the launch of GPT-5, the drop in ChatGPT referral traffic, and the rollout of Google’s first spam update of 2025, and so we’re back with a Digital Digest full all the latest news and fresh innovations that you need to know about.
ChatGTP 5 Is Here
GPT-5 has officially been released by OpenAI.
The latest ChatGPT model, hailed as the ‘smartest, fastest, most useful’ one yet, has fully finished rolling out across all tiers. As the first new model to be released in over a year, the launch of GPT-5 has been highly anticipated, but is it the breakthrough that OpenAI promised it would be?
GPT-5 has replaced GPT-4 as the default model in ChatGPT, and the reviews widely consider it to be an improvement on its predecessor. It has been deemed much faster than the models that have come before it and delivers quicker responses without compromising on the accuracy and usefulness of its output. It has also been programmed to be less ‘sycophantic’ in line with a new form of safety training, which is a welcome development as this over-compliance could often lead to incorrect answers and ‘hallucinations’ with previous models.
Google’s August Spam Update
Google’s latest update, launched on August 26th, is its first spam update of 2025.
Last month, Google confirmed that it had begun rolling out its latest update on August 26th and that this could take a few weeks to complete. Whilst there have been countless core updates shaking things up all year, this is the first Google Spam Update of 2025, and the first one since Google updated its Quality Rater Guidelines earlier this year.
This update is described by Google as being a ‘normal spam update’ that applies globally and will roll out for all languages and locations. No further information was given regarding what type of spam this update is currently targeting, but since Google defines spam as content or techniques used to deceive users or search engines it is safe to assume that any significant ranking or traffic changes are a result of this latest update.
Are AI Results Verified?
It’s a question that is still being asked nearly two years into the AI boom.
How many times have you read an AI Overview and come across information that is completely false, or asked ChatGPT a question only for it to generate an incorrect answer? In AI terminology, the term for this error is a ‘hallucination’; in human terms, we call this a mistake.
No matter what it is called, the fact remains that AI models are able to produce false results and make unsupported claims. A recent Salesforce AI Research study found that as many as one in three AI-generated answers contain information that is not backed up by reliable sources, and that popular AI tools with millions of global users are consistently delivering unsupported statements. Bing Chat failed to back up 23% of its statements with reliable sources throughout the study, a stat which grew to 47% for OpenAI’s GPT 4.5, and soared to 97.5% for Perplexity’s deep research tool.
ChatGPT’s Referral Traffic Plummets
ChatGPT referral traffic has decreased by 52% since 21st July.
Since 21st July, referral traffic originating from ChatGPT has dropped by an astounding 52%. Recent analysis of ChatGPT citations shows that, whilst global ChatGPT usage continues to grow, clicks from the platform though to the websites that it cites have been steadily declining.
Yet a handful of websites have come out of this dip with a much larger share of referral traffic than before and nearly 1 in 5 ChatGPT citations now go to just three sites: Wikipedia, Reddit and TechRadar. Between them these sites now account for 22% of all citations, and Reddit citations alone are up by 87% since July, while Wikipedia has hit historic highs with an increase of 62% in the same time period. This change has not been caused by GPT-5, as referral patterns started to change long before the new model launched on August 7th, and this suggests that the traffic dip stems instead from an ongoing OpenAI experiment in how the platform finds and cites online information.
The Illusion of Information
OpenAI advancements seek to minimise AI answers that aim to please rather than inform.
Many users that turn to AI for informational purposes do so for its conversational tone and perceived personability, but this politeness poses real challenges for public access to reliable information. Most AI models are easily influenced to answer a user’s question in a particular way based upon its phrasing and context, and these platforms would sooner produce incorrect results than admit they couldn’t find the right information, often giving speculative answers or incorrectly citing sources in order to answer questions.
One of the main goals that OpenAI hoped to achieve with the launch of GPT-5 was an increased ability to follow instructions and deliver clear information whilst minimising the models’ sycophantic way of communicating. Already, instances of sycophancy influencing the accuracy of results have reduced by 75% for the paid version of GPT-5 and 69% for the free version, and the hope is that this improvement will also lead to fewer factual errors and reduced hallucinations across the platform.
The Digital Digest
We publish updates just like this one every single month, so come back soon to stay on top of all the latest developments from across the digital marketing industry.
Our team is also on hand to help you address any of the marketing challenges raised in this Digital Digest or any of our previous editions, so feel free to reach out today.
