The AI Effect

Over the past two years it has become increasingly noticeable that writers on our retreats fall into three groups when it comes to the use of AI in their writing. The first group are generally sceptical, feel it is cheating, and avoid its use (although some in this group are simply not interested in exploring it). The second group, in some ways the binary opposite of the first group, use it for everything: research, editing, grammar and spell-checking, proofing, and anything else it can do - including occasional drafting of chapters and more. The third group, sitting somewhere in the middle, use it for very specific tasks such as reviewing a draft for structural issues or executing grammar-checks. The middle group, above all, seem to understand the abilities and limitations of AI extremely well - generally noting that the instruction you give the agent has a significant effect on the output you get back. Either side of this group sit writers with strong views on either side of the AI debate.

The issue of AI has interested me personally for a long time, and beyond the realms of writing. But today I encountered my own latest challenge on the topic and it has led to a fresh review of the capabilities and limitations as I understand them. The issue that arose today concerns the visibility of our website in the AI age: Google algorithms used to love the fact that our site is called A Place to Write, it offers A Place to Write, and people interested in our offer frequently search for A Place to Write. This morning, however, when reviewing our website statistics, a new trend emerged: we are being increasingly found - or not found - by AI agents, such as ChatGTP, which are reporting back to a user on a question they have asked about writing retreats. In this format we are being recognised by the AI agents but without the clarity that Google used to provide. As critics of AI have been predicting for some time now, the extended responses given by AI in response to a query are seen as a replacement for the need to go and explore a website close up; if you are particularly interested in the AI response then you might visit various websites it mentions but more often than not the AI summary is used as a guide to which specific site is worth bothering with. If the AI is not reporting a website accurately, it can be missed by the reader. And this is the situation I discovered this morning.

In the case of A Place to Write, I found that the AI agents recognise the excellent reviews we get (currently 5.0 stars on Google and 5 bubbles on TripAdvisor) and have gathered some key information about our retreats. However, in providing summaries for human researchers, there were two consistent errors: firstly, the agents reported that our location in the North Wales countryside meant we were inaccessible for anyone without a car. Wrong! We offer complimentary pick up from the nearest mainline train station and the bus service in North Wales is superb (should you wish to travel to us by that means). Secondly, AI summaries stated that the small group (6-7 people) was a negative - meaning we were often sold out. Wrong again! It is true that we had a fantastic spring and we did sell out some of our retreats at that time - but we actually have up to 9 rooms for writers and sell-outs are still rare. The small group is actually a huge part of the magic of our retreats, as guests consistently report. AI has therefore been misrepresenting two features of our retreats for some time now!

A third point which AI arguably also gets wrong is the pricing: it claims our bespoke, personalised, small group retreats will make us more expensive than larger, self-catering retreats. At the last check, our untutored retreat price for a superior room was highly competitive and, if you acknowledge the level of hospitality and support mentioned in reviews, somewhat cheap at the price. Yet this blog was not intended as a rant about AI but an update: I have now added some software code to the website which, apparently, will help AI get its summaries more accurate in future. Whether it will work or not, I don't know. Time will tell. But it was a salutary lesson for me that no matter how I might feel about the benefits and costs of AI, it is rapidly changing the world we live in and there is an imperative to respond. Even if, in AI, everything is still just reduced to a number and a set of probabilities (yes, that is literally all it is!) it is a mathematical force which, given the ridiculously massive investment it has generated, has capacity for both good and bad.

But I still prefer meeting, talking, socialising, debating, educating and learning from humans...

Till next time, happy writing,

Nic