AI Can’t Practice Law

by:

Joe Patti

Entertainment lawyer Gordon Firemark made a blog post recently warning people not to ask AI for legal advice. As you might imagine, he cited some examples of how AI had made mistakes, gotten the law wrong, and generated language that put all the risk on the client rather than the supplier of the service.

He also noted that AI tends to validate the perspective of the person asking the question. He related the story of a client who wrote a strongly worded letter based on AI advice which created an antagonistic relationship with another party which resulted in a prolonged conflict between them.

He said the whole issue could have been solved with a minor revision in contract language and a conversation.

But even more importantly, Firemark says, when you ask AI for advice or to help you produce something, you are making assumptions about what the solution is. One example he gives is asking AI to generate an independent contractor document when the legally the relationship is employer-employee.

Likewise if you aren’t familiar with the difference between copyrights and trademarks, you may end up insufficiently protecting your corporate identity. You may incorporate your business using the wrong/disadvantageous structure.

As much as you may wish to avoid the cost of working with a lawyer, they are still able to evaluate your goals and the context of your requests better than AI. They can also help you recognize that what you think is important isn’t really relevant and that you have overlooked some crucial arrangements.

If you’re facing a legal issue, resist the temptation to ask AI what you should do.

Don’t ask it what agreement you need.

Don’t ask it what legal arguments to make.

Don’t ask it to draft the first version of an important legal document.

Bring your problem to your lawyer instead.

Bring the facts.

Bring your goals.

Bring your concerns.

[…]

Because the value isn’t in the document.

The value is in the thinking that produces the document.

Not As Many Agree With You As You Think

by:

Joe Patti

Last month Seth Godin linked to a seven question quiz based on data from the General Social Survey.

Apparently, the most your responses can align with those of others in the United States is 7%.

Godin scored 4%. My score was around 1%.

So when you believe that millions of people think like you do, you are technically correct in that at most ~24 million people out of 342.6 million in the US hold the exact same set of beliefs as you.

But also there are even more millions who don’t.

I have been writing a bit about marketing to different audiences the past few weeks. In the context of this data it makes me wonder how valid the exercise of creating hypothetical profiles of audiences actually is.

If at most 7% of the community you are trying to reach is going to perfectly share the beliefs of others in the community, you are really angling for a small segment of the community.

The percentage is guaranteed to be smaller than 7% because at least a few percent will be knocked off when people who don’t like or feel comfortable with whatever you are offering self-select out.

Obviously the remaining 2%-4% (which may still be an overestimate) don’t have to be perfectly in-synch with one another to enjoy experiences your organization offers.

Ultimately, participation in arts and culture may indeed have niche appeal because that is the best you can hope for.

Though again, 2% of the US population is about 7 million people which isn’t a small number, except in relation to the whole nation.

Tips For Generating AI Messaging For Different Audiences

by:

Joe Patti

Ceci Dadisman posted some guidance on LinkedIn for arts marketers using Ai to write promotional copy. She talks about creating different messaging for each segment of an audience. In this case, Insiders, Recent Buyers, Lapsed Buyers, and Cold Prospects.

The basic approach is the AI agent is writing the four different sets of copy you would write if you didn’t have the time. This is different from having AI writing copy for you because you are taking responsibility to review the copy and revise parameters so that there is basically no commonalities between the message for Insiders and Cold Prospects.

Ceci walks through the steps of writing a comprehensive prompt that ensures that messaging is crafted specifically for each audience. This requires providing details about the audience, what they know, what they don’t know, the voice to use, and most importantly; what concepts, words and jargony terms not to use.

For example from her prompt instructions:

When rewriting, do:

-Lead with what the reader will experience, not what we are offering
-Use “you” more than “we”
-Define any word a smart friend outside our industry wouldn’t recognize
-Keep sentences short. Sixth-grade reading level or below.

When rewriting, never:

-Use the words: [list of terms — e.g. “transcendent,” “iconic,” “immortal,” “unforgettable”]
-Assume the reader knows any staff member’s name
-Reference internal season structure or programming language without translation
-Add urgency the original draft didn’t have

She includes tons more insight and instructions, including how to test and evaluate what AI produces against itself and content a human has written.

Google Wants To Be The Destination Over Your Website

by:

Joe Patti

Seth Godin asked Google’s AI service Gemini about how Google has undermined blog and creator generated content.

Godin gave the AI credit for being honest about its corporate masters. It reported that shutting down Google Reader in 2013 meant that blog content was no longer easily accessible, breaking down the relationship between creators and readers, and shifting people to algorithm driven social media.

Then in 2023-2024 algorithm updates saw independent publishers see organic search visibility plunge 40% to 90% overnight..

Finally Gemini noted that the inclusion of AI overviews in search results has seen independent publishers and tech sites lose up to 58% of their traffic since early 2024.

This latest phase is also impacting arts organizations as well. Because Google is scraping sites for content to provide answers, a lot of arts organizations have seen their visitation numbers drop precipitously because people don’t have a reason to visit.

Google actively scrapes blog data to answer questions directly on the results page. This creates a “dead end” where over half of all web searches finish without a single user clicking through to an external website. 

[…]

 When I fail to fetch an exact quote from an independent blog because my search component strips formatting and acts like a centralized answer engine, it functions exactly like the broader systemic bias you are pointing out. 

The modern internet has been built to turn search engines from gateways into destinations, starving independent creators of the clicks, traffic, and revenue required to sustain their work.

While Gemini was responding to Godin’s query about the impact on blogs and independent creators, its answer sheds light on the reason arts organization websites aren’t getting the traffic they once did. Google’s goal is to have the information they provide be where the visitor stops.

I performed a search asking what was happening at my theater and the response listed the next 6-8 chronological events. But there is no mention of tickets for our upcoming season going on sale soon which is something visitors would learn more about if they visited. I performed similar searches on peer organizations who had also recently announced seasons and there was no mention of their seasons being on sale in the results.

For some, there was no AI summary either which means people would have to click through to the website. I am wondering if it might be beneficial to design a website so that the content can’t be easily scraped and summarized. Though I also wonder if those sites might be demoted in future search results.

The searches I did specifically asked “What is happening at X,” where I mentioned the venue by name. But if I asked what is happening at theaters/performing arts centers in a city or region, those that can’t be easily scraped for data may not appear prominently in the results.