Working notes on AI marketing for local businesses, written from live client engagements rather than from vendor press releases. Where the evidence is thin, we say the evidence is thin.
The Journal is where My Local Everything publishes what it is actually learning from live client work on AI marketing — GEO, AEO, AI advertising, agents and automation. In short: observed findings, stated uncertainty, and corrections published rather than quietly edited.
The current article works through the difference between paid AI advertising and GEO, and why running only one of them is the expensive mistake.
ChatGPT is becoming an ad surface. It is also, separately, an answer surface — and the businesses being named in those answers mostly are not paying for it. The honest difference between the two, what each actually costs over time, and why running only one of them is the expensive mistake.
Read the articleMost agency blogs exist to catch search traffic and say nothing. This one exists because the AI marketing field is moving fast enough that a good deal of confident public advice is wrong, and we would rather put our reasoning where clients can check it.
So: things we have actually observed in live engagements, distinctions the market keeps collapsing, and the occasional correction when we get something wrong. Where the evidence is thin we will say the evidence is thin.
The Journal covers the parts of AI marketing where the public advice is thinnest and the consequences of being wrong are largest.
AEO and GEO are not the same discipline. Paid AI advertising and GEO are not alternatives. An AI agent and a chatbot are not the same product. Each of those confusions is currently costing somebody money, usually because a vendor found it convenient not to explain the difference.
We run live client engagements across ten industries, and the transcripts, the answer-engine tests and the reporting tell us things that are not in anyone's press release. When what we see contradicts the received wisdom, that is worth writing down.
Attribution inside an AI conversation is genuinely hard. Model outputs vary run to run. A confident-looking number built on assumptions is worse than an honest gap, and a great deal of AI-marketing reporting is currently the former.
Consistent facts across every directory. Content a crawler can read without executing JavaScript. Specific numbers instead of adjectives. None of it is exciting and all of it outperforms the tactics that get written about more.
This field is moving fast enough that some of what we publish will not survive contact with next year.
When an article turns out to be wrong, we correct it visibly and date the correction rather than quietly editing the page and hoping nobody noticed. When a figure comes from someone else's research, we link the source and state its limitations — including when the headline number is an upper bound rather than an average, which is the most common way a genuinely useful study gets misquoted.
When we are describing something we have done for a client, we say so. When we are describing something we expect to happen, we label it as a prediction. And where a claim would need client data we do not have permission to publish, we leave it out rather than gesture vaguely at it.
None of that is remarkable. It is simply the standard we would want from anyone we were taking advice from, and it is unusually rare in marketing writing right now — which is most of the reason this section exists at all.
That is usually a better conversation than a sales call anyway.