AI for Marketing Professionals — American Marketing Association
Generative AI took the world by storm in fall of 2022, we started experimenting at Sund + Co, and soon we were users! This is a summary of the presentation we gave to the Digital Marketing section of the Cincinnati American Marketing Association on 21 June 2023.
AI for Marketing Professionals
What it is and how to prompt it
Mark Sund, Studio Director, Sund + Co
Sarah Crawford, Creative Director, Sund + Co
Dino Pelle, Digital Director, Engage321
21 June 2023
AI is Like…
Working with an intern or entry-level employee, giving someone too tight a deadline, working with a natural liar, working with an incurable pleaser (it won’t challenge you, fulfills every request as if it were sensible), working with a brainstorming partner, and managing a team of employees.
We’re giving away the conclusions up front!
Things to Think About
What this banner AI-generated?
Was this talk outline AI-generated?
2016 State of the Art
Tay was developed by Microsoft, given a Twitter account, and released to the public, where the rest of its training was crowd-sourced in real time. Tay’s goal, like all business ventures, was to keep users engaged so they didn’t leave. Tay quickly learned that misogyny, racism, violence, and sexuality kept conversations going! Microsoft took away Tay’s Twitter account less than 24 hours later.
Here’s a conversation I had with Tay.
What Changed in 2022?
First there was evaluative AI.
Then came 1:1 A.
Now we have generative AI.
AI Tools and Techniques
Fundamental AI Tools – Design
Dall-E
Midjourney
Stable Diffusion
Runway
Adobe Creative Cloud – pre-existed the 2022 explosion with content-aware fill and continues to embed more AI
Live Demos and Examples – Design
We started with design, instead of verbal, since it’s quicker and easier to see how to revise your prompts to get to a good final product.
Visual and verbal AIs make similar types of mistakes and benefit from similar types of prompting conversations.
Fundamental AI Tools – Writing
ChatGPT 4 (pay the $20/mo for version 4 – slower but better)
New Bing
Google Bard
Perplexity – great for doing research
Predictive text – pre-existed the 2022 explosion – AI writing tools are basically predictive text on steroids
Live Demos and Examples – Writing
Here we showed a host of examples for writing articles and social media posts, optimizing for search engines, creating content like recipes, different tones of voice, posts about the advantages of shopping for clients’ products on their website, on Amazon, and at local stores – all without having to shift mental gears when writing; creating marketing plans and calendars; and more.
AI Tools for Marketing – Articles and SEO
ChatGPT 4
Copy.ai
Jasper – pre-existed the 2022 explosion
WriteSonic – pre-existed the 2022 explosion
AI Tools for Marketing – Websites
Mixo and Durable are interesting but not professional grade yet, and you have to host on their minimally tested platforms. Squarespace and Elementor (and other WordPress plugins) are existing professional grade website platforms that are incorporating AI into their tool kits – a better bet.
AI Tools for Marketing – Campaigns
Keeping your projects and brand kits organized has been a theme today. Jasper, Microsoft Designer, and Adobe Express are new-ish AI platforms that try but don’t do it real well (yet). ChatGPT works if you’re careful with your conversation threads.
Canva, Hubspot, and others are pre-existing campaign platforms that are incorporating AI. As with websites, the best bet at present is to use existing platforms that are adding AI to their tool belts.
AI Training, Legal Aspects, and Risks for Companies
Training and Biases
The Washington Post got hold of the list of websites used to train an AI model. Some far right, some far left, some that have been shut down for legal violations. OTOH why shouldn’t an AI model know about these instead of hiding its head in the sand. Religion and philosophy mostly Western. Nothing scandalous exactly but it’ll make you go hmmm. A must read…
https://www.washingtonpost.com/technology/interactive/2023/ai-chatbot-learning/
Live Demos and Examples – Training and Biases
Skipped the live demos for time, but described how we had ChatGPT write jokes and it referred to animals as “he/him”. Same bias we tend to have too if we’re not careful and for the same reasons.
In fact, AI models tend to amplify the biases they’re trained on – they’re incurable pleasers and that’s how they please the most people.
Copyright
For creators whose work the AIs train on, Mark’s position is AIs are not people, so the concept of fair use doesn’t apply – we value artists learning from talented creators and reviewers quoting from their works. But profit-making companies shouldn’t have human rights and should pay creators for commercial use of their work.
For users who use AI, can they claim copyright on the output? It depends on how the courts rule about copyright cases for creators, above, and it depends on how much original, skilled, human work went in to the final product.
Business Risks 1
Getting the facts wrong, low-quality writing / design / code, sounding average or like a parody of what you intend. These are risks when people do work too. Everything’s faster with AI, but quality control isn’t any faster.
Another business risk didn’t make this slide: depending on court rulings, the AIs you’re using could be violating others’ copyrights.
Business Risks 2
Whatever you enter into an AI, it can be trained on – absorbed into the model and presented to other users, either associated with your company or presented as established fact.
Some AI models have a do-not-train option, which seems great, but they also build in inconveniences, it’s unclear how well it’ll be enforced, it’s already nearly impossible to keep track of this for embedded AI and especially when the same model is used across multiple platforms. On top of those, do-not-train doesn’t prohibit the company from using or selling the info you enter.
Take-Aways
In order from sloppiest to most careful, keep in mind what you’re doing…
Brainstorming (internal)
Research (internal)
High volume, good enough (output)
Low volume, high quality (output)
Curate and Iterate
– Terry Campbell, CPG photographer mentioned during live demos
Your weird + AI’s magic
– Molly Mahoney, The Prepared Performer
It’s not a prompt, it’s a process
– We said that
Curate and iterate and document – for copyright purposes
No proprietary info – don’t feed this to the models
Plan how to organize clients and projects. Each conversation in ChapGPT has its own memory; some tools have well-developed brand kits.
Play around first
Give it the facts: company, products, services, positioning, etc
Tell it the style and audience
Prompt, revise, repeat
Take over for human editing
AI is the best / worst employee you ever had
We’ve come full circle. Working with AIs is similar to working with people in these situations.
If weren’t a manager before, congratulations, you are now!
Prompt well and prosper
Thank you
To the AMA, Cincinnati Digital Marketing Community, and Betty Rodriguez-Hakes!
Q & A
Had just a couple minutes for discussion!
Marketing CASE STUDIES
We do advertising, market research, and lead generation for a wide variety of B2B and B2C brands.
Need expertise to acquire clients and generate insights?
