People often claim that prompt tracking cannot work because every prompt is unique.
We ran the numbers and can confidently say that this concern is not justified.
This conclusion is based on two experiments:
1️⃣ We analysed two sets of prompts that were written by Rand Fishkin’s followers.
2️⃣ We created our own prompt sets where we changed a set of base prompts by the minimal possible amount as often as possible without changing the intent
Here are the facts:
• While every human-written prompt was unique, 90% fell into a bucket of similarity where the likelihood of a brand being mentioned in the LLM answer does not really change.
• The style of the prompt matters a lot. Asking for the best or a list can greatly increase the number of mentioned brands. Giving the LLM a role (“you are an expert on SEO”) leads to fewer brand mentions.
• Both top and bottom of funnel prompts are robust against wording changes. Mid of funnel prompts however are much more sensitive. Small variations can quickly surface different brands in the answers.
• In ChatGPT and Perplexity, constraints reduce the number of brands shown. In Gemini and Google AI Overviews, constraints actually increased the number of brands. Potentially by triggering additional fanout queries.
• The length does not matter. As long as the intent stays the same, conversational fillers words do not significantly impact AI answers.
What does this mean for me?
• Do not obsess over exact prompts wordings. Focus on topic, intent, funnel stage, and context.
• Consider to be more granular in the mid of funnel prompts you are tracking. Here every prompt variation is most likely to surface additional brands, sources, and insights.
The ' social media manager ' has developed into one of the wildest job descriptions in the modern marketing team.
I just saw one posted
- strategy
- creation
- moderation
- trend monitoring
- relationship management
- analytics
Drive revenue and be on-site everyday for 45k.
@BradMichelson This is the part they don't tell you. It usually involves getting a massive people to do a behavior change, and people don't like change.
WordPress is SLOW (usually) but... it doesn't have to be.
After working with WordPress for 15+ years as the plugin developer - I've picked a thing or two (or 30) about WordPress performance.
And it's all beyond the obvious:
"Install caching plugin"
and
"Buy a $300/mo hosting" ( this is bullshit btw. if you're paying more than $20-50/mo you're being ripped hard)
So I wrote a piece about it. It's very low-level and technical mind you, but maybe you'll find something useful there.
https://t.co/yCbjWpr7cQ
Ai design tools are cool and all, but have you ever just put text on a picture in Instagram, exported it and run it as an ad.
I've got a couple that are absolutely slaying 'brand' creative
I've been switching my defined AI workflows to be agents. Aside from some hassle setting them up, the main thing I've gained is significantly more token usage 😆
@realLiamGill Unfortunately, the length of the sales cycle and the unreasonable demands made in these tender contracts make it nearly impossible for startups to grab any of that business