Why you shouldn’t use GPT to write outreach
Sure generative AI can imitate personalised messages but we’re not convinced that it is the future of TA.
The rise of generative AI tools such as GPT and Bard has led to many people applying the technology to mundane and repetitive tasks. Personalised outreach is one of these big tasks for recruiters since many wish to personalise messages but lack the time and inspiration to write them.
In 2022 we wrote north of 6,000 uniquely individual outreach messages to candidates and we’ve learnt a lot. The biggest thing is that personalisation is super effective, and you just can’t fake it – sorry GPT.
Whilst generative AI can do some pretty amazing things, the fact is generative AI has far to go.
One of the things that you’ll immediately notice is GPT messages very often lack a tone and voice, making it sound highly generic and disingenuous. Most of these messages are easily identifiable as being generative AI for that simple reason.
This means that if you’re generating individually personalised messages for candidates you’ll be spending a lot of time editing each one to match your tone. Not only that, but generative AI tools are also at risk of hallucinations, something that occurs when generative AI fabricates information. So you’ll need to go and verify all the information in the message is correct.
Once you’ve edited the tone and verified the content of the message it would have been just as fast to write out your own message, and the end results would have been better.
The Same Message
One of the other big issues is that candidate’s inboxes are flooded. Vincent Chi says that a Software Engineer receives at least 37 InMails a month. Candidate’s inboxes are being flooded by “personalised” messages that all look the same, making it harder than ever to cut through.
Because everyone is accessing the same AI model when they query the software there is a high risk of everyone getting the same or similar output. This exponentially increases when you consider that most people are using the same prompt: “send a personalised outreach message to this candidate.”
Instead of standing out from the crowd with your “personalised” message, you’re actually joining a long list of messages with the exact same personalisation. On the plus side, this makes it easier for those who are actually personalising their messages as their genuine messages stand out.
All of this is really adding up to one thing: candidates know.
Candidates knowing that you have not actually personalised your messages, and are pretending to have done so leaves a sour taste. The messages are dishonest, you’re trying to fool candidates into thinking that you’ve spent time and effort on a message to you. The almost transactional nature of spending time personalising a message with the expectation that a candidate will spend the same time replying to you is removed. It is almost better to just use a traditional template because at least you’re not being dishonest.
We talk about how outreach messages are often not considered to be a part of employer branding, even though they are the candidate’s first impression of a company. Between the clear lack of effort, similarity, potentially false statements and dishonesty generative AI messages are definitely not good employer branding.
Ultimately, with some human editing, many of these generative AI messages could be fine. But, at what point where you are editing every AI-crafted outreach message should you just do it yourself from scratch?
So, if GPT isn’t the answer, what is?
We think it’s good hand-typing and improving your work. There are plenty of resources on writing effective outreach available on our website including:
- Our Hot Source podcast series
- Writing Candidate-Persona messages
- 8 basics of outbound messaging
- Unconventional outreach strategies