Zhang Pinpin · Hands-On AI Coach
← Back to home

GEO Is Multiplication, Not Addition: Product Substance Is the 1

Long-Form Video · EP0050 June 24, 2026 5:57
What this episode covers

This episode may offend a lot of people. We're going to talk about GEO, or "Generative Engine Optimization."

Why offend people? Because GEO has become one of the best sources of business profit for a lot of brand marketing and advertising agencies right now. It isn't that hard to do and the investment isn't high, yet brand demand is substantial and brands are especially willing to spend marketing budget on it. Today, though, I want to voice the dissent: GEO is nowhere near as good as it looks, and not every business and not every industry is suited to it.

Companion files · Drop them into Claude Code
GEO Playbook + Pitfall Checklist HTML 38 KB
Everything in this episode about GEO (Generative Engine Optimization) organized into a one-page checklist you can follow. The top half is the four-step how-to (work out which model your customers use → have AI generate the questions they'd ask and see which competitors show up → ask the model for its citations → write your brand information as original content AI likes and publish it to those sources). The bottom half is the when-not-to-do-it checklist and judgment framework (GEO is multiplication, not addition; product substance is the 1; which businesses should and shouldn't do it; AI head-to-heads filter out every bit of marketing language). One page, no detours.

This episode may offend a lot of people. We’re going to talk about GEO, or “Generative Engine Optimization.”

Why offend people? Because GEO has become one of the best sources of business profit for a lot of brand marketing and advertising agencies right now. It isn’t that hard to do and the investment isn’t high, yet brand demand is substantial and brands are especially willing to spend marketing budget on it. Today, though, I want to voice the dissent: GEO is nowhere near as good as it looks, and not every business and not every industry is suited to it.

First, exactly how GEO is done

GEO stands for Generative Engine Optimization. Here’s the idea. In LLMs like Doubao and DeepSeek, people ask questions and the model writes the answer—and those answers mention brands and products along the way. GEO is the work of getting more information that favors your brand and product into what the model writes.

Roughly, it goes in these steps:

  1. Work out who you’re selling to and which model they mostly use. If I’m selling to a particular audience in China, the generative engine they use most is Doubao.
  2. Use AI to generate a set of questions. Have the model itself hypothesize: if my customer segment went to Doubao, what would they ask? Feed those questions to Doubao and look at which industry competitors and which industry news its answers pull in—that’s round one.
  3. Ask it for its citations. Then you can ask Doubao for the sources behind that generated content: “Where did you get this information?” It will very likely tell you—this came from Zhihu, that came from some official site, that other one came from a video on some video platform. It hands you the sources.
  4. Write your brand and product information as original, distinctive content AI likes, and publish it on those source platforms. Do enough of it on the platforms it cites and that content gets picked up by search engines; when a user asks, it shows up in the generated answer.

It sounds like a tidy, self-reinforcing machine. And that’s exactly where the problem is.

It’s multiplication, not addition

The first problem to watch: GEO raises the exposure of your brand and product information, but it does not determine how that information gets presented and used.

And in today’s cutthroat consumer market, comparing products takes almost no effort. So if your brand and your product are weak on substance, getting presented may not be an advantage at all. It may be a disaster.

There’s also an argument that GEO is overrated, because the model companies control what gets presented. Say OpenAI’s GPT someday puts an explicit price on placement and folds advertising into its generated answers the way the client wants. It will then find some way to control how often—and in what form—a brand’s own GEO content shows up.

So my core view is this: GEO is multiplication, not addition.

Only when the product is genuinely strong and the brand is being discussed and positively mentioned across the internet does it become an amplifier for that brand and product. It’s the pile of zeros after the 1—but it is not the 1.

That said, GEO isn’t useless. People use search engines less and less, so optimizing this kind of content really does raise brand visibility and convert traffic better. And to a degree it counts as third-party endorsement rather than the brand talking about itself, which makes it carry more weight than other channels. But as I said: ranking alone gets you nothing, and weak product substance turns visibility against you. Get the product and the brand right first, then talk about GEO.

GEO can only raise the probability that an LLM “gets hold of” your brand and product information. It can’t determine how the LLM “uses” that information.

Evidence: I picked a tablet with Claude Code

Take the tablet I bought a while back as an example.

I used AI to find my “chosen one” pad. The prompt I gave Claude Code: released after February 2026, 12 to 14 inches, Android tablet (I don’t like iPads, and I’m not that keen on HarmonyOS either). I also listed a few things I cared about: as thin and light as possible, the best screen quality I could get, a high screen-to-body ratio (i.e. narrow bezels), ideally equal bezels on all four sides (I’m obsessive about this, so narrow and equal on all four sides scores highest).

Working from those requirements, it ran deep research and a head-to-head comparison. Everything it gathered along the way came from channels it judged more credible. It listed 20-plus tablets, generated a full spec comparison, and highlighted the best value in each row—highest resolution, highest screen PPI sharpness, highest refresh rate, highest brightness, all compared.

Then it picked for me based on my requirements: on sharpness, the vivo Pad6 Pro is the best choice—a superb 4K panel hitting 347 PPI, extremely fine. But on weight, the Honor MagicPad 3 Pro is only 450 grams, far below comparable competitors, and it’s also the thinnest, has the largest screen-to-body ratio, and has equal, very narrow bezels on all four sides.

Based on the characteristics I cared about most, it produced a weighted-average scoring table. In the end, the Honor MagicPad 3 Pro (12.3-inch) took the highest weighted score. I love it now that I have it—genuinely thin, equal bezels on all four sides, no hole punch, and remarkably light: about the weight of two phones.

In AI head-to-heads, marketing copy gets stripped bare

So was GEO important in that process? Yes. And not just GEO—SEO matters too. I asked about models released after February 2026, which means the LLM’s own knowledge base may not carry the details. The agent has to run the web search a human would have run, then combine what it finds with its internal knowledge and hand me an advisory-style report with a conclusion. Skip GEO and SEO, and your product’s specs could never have made it into that comparison.

But equally—even if you’ve done SEO and GEO in full, if the product substance isn’t there, if the product itself doesn’t hit consumers’ core needs, then having it presented does nothing.

And you may have noticed: once AI had chewed through all that information, I never saw a brand’s precious value proposition, its slogan, or anything about “our national champion, far ahead of the pack.” None of it survived. All I saw was one question answered: why does this product fit my needs?

In an era where AI makes the decisions, marketing copy is stripped bare, and product substance is the trump card.

So: how much GEO really does for you, which brands should use it, which shouldn’t—I suspect you already have your own answer. Let’s talk in the comments. See you next episode.

It's the pile of zeros after the 1—but it is not the 1.