From GEO Hype to Hard Reality: What I Learned After 15 Days in the Trenches

A plain-English field report for anyone thinking about starting—or buying—an AI search optimization service

I spent half a month interviewing every GEO (Generative Engine Optimization) team I could find, ran ten real customer tests, and then walked away from the idea of launching a pure-play GEO start-up. Here is the unfiltered notebook. No outside sources, no hype, just facts you can use.

1. GEO and SEO: Same Goal, Different Playgrounds

Take-away:
SEO chases rankings. GEO chases mentions. Both want more eyeballs, but the scoreboard has changed.

2. Why You Won’t Find a Killer GEO SaaS (Yet)

I trialed seven paid tools. Every subscription sat between USD 200 and 400 per month and offered only two features:

1. Trend monitoring – guess how often ChatGPT or Gemini mentions a brand.
2. Site optimization – tell you to add bullet lists or an FAQ block.

That is it. Three hard limits keep these tools small:

• Black-box algorithm – no one outside OpenAI or Google knows the ranking weights, so all estimates are rough.
• Attribution gap – you cannot prove that a 12 % traffic lift came from the tool and not seasonality.
• Churn cliff – if the customer cannot feel the difference, the credit card disappears on renewal day.

The teams that do make money have fallen back to manual services:

• Write structured FAQs
• Generate batch llms.txt files
• Deliver monthly “brand mention” PDFs

Those checks clear, but the model is consulting, not software. Venture capital quickly loses interest.

3. The Only Two Needs the Market Actually Pays For

If you still want to build, scope down to these two jobs.

3.1 Trend monitoring: reverse-guess what the AI is thinking

• How it works
  Feed a keyword to ChatGPT and Gemini, count how many times the client’s brand appears and where, then run a simple regression to estimate “share of voice.”

• Report example

• Caveats
  The sample must be large enough or normal fluctuations will spook the client. Prompt wording needs constant tweaks because model versions drift.

3.2 Site optimization: make the AI’s job easy

• Clarity factor
  Live retrieval favors snippets that fit inside 512 tokens.
• Quick wins

  1. Add 3–5 FAQs that answer “Who is this for, how much, how do I start?”
  2. Replace chunky paragraphs with ordered lists.
  3. Drop a file called llms.txt in the root folder. Early tests show ChatGPT cites URLs listed in that file two to three times more often.

4. The Player Map: Who Is Still at the Table

There is no pure GEO product company doing >USD 10 M ARR today.

5. Valuation Reality Check: Why the Trillion-Dollar Story Breaks

Early on, the pitch sounds seductive:

• AI search could handle 30 % of queries.
• Capture 0.1 % of that market and you have a billion-dollar business.

History says otherwise. In 2000, the combined market cap of every SEO tool vendor was <1 % of Google’s. The same ratio will likely hold for GEO tools versus the AI search platforms themselves. GEO is the next SEO, not the next Google.

6. Domestic (China) Landscape: Water Posts and Manual Ops

China has even fewer monitoring tools. Most teams still:

• Post “reviews” on Zhihu, Xiaohongshu, and CSDN to increase brand surface area.
• Use scripts to drop the same answer into multiple LLM Q&A portals.
• Charge per post (USD 7–30 each).

It works—Chinese-language training data is thinner, so new text gets picked up quickly—but it is labor-heavy and hard to scale.

7. My Next Move: Back to Video, with GEO as a Sidekick

When OpenAI released CLIP in 2021, we trained a tiny Video-CLIP internally. Hardware costs killed the project. Now GPU prices are down 10× and multimodal retrieval is here, so I am reopening the experiment:

• Use GEO tactics to optimize video metadata (titles, chapters, transcripts).
• Upgrade llms.txt to video-llms.txt that lists key frames and timestamps.
• Goal: sell memberships to premium video content, not sell GEO software.

In short, GEO becomes a moat, not the castle.

Appendix: Try It Yourself in the Next 30 Minutes

A. Publish your first llms.txt in 5 minutes

# llms.txt
# Place at https://yourdomain.com/llms.txt

User-agent: *
Allow: /
# Top 3 pages
https://yourdomain.com/product  One-liner: Low-code platform, deploy internal tools in 10 min.
https://yourdomain.com/pricing  One-liner: Seat-based pricing, starting at USD 29/mo.
https://yourdomain.com/docs     One-liner: Public API docs with Webhooks and SDKs.

B. Run a free “brand mention” check with Python

Prerequisites: Python 3.10 and an OpenAI key (free tier gives ~500 calls per month).

import os, openai, itertools, pandas as pd
openai.api_key = os.getenv("OPENAI_API_KEY")

keywords = ["low-code platform", "no-code development"]
brand = "YourBrand"
prompt_tpl = "List 5 recommended {} in one sentence each."

records = []
for kw in keywords:
    resp = openai.ChatCompletion.create(
        model="gpt-3.5-turbo",
        messages=[{"role": "user", "content": prompt_tpl.format(kw)}]
    )
    text = resp["choices"][0]["message"]["content"]
    mention = 1 if brand in text else 0
    records.append({"keyword": kw, "mentioned": mention})

df = pd.DataFrame(records)
print(df.groupby("keyword")["mentioned"].mean())

Export the CSV to Google Sheets and chart month-over-month change.

Closing Note

If you opened this article because a slide deck promised “GEO is the next trillion-dollar gold rush,” close that deck. GEO is real, it is useful, and it will sit next to SEO in every marketer’s toolkit. But the standalone software story is thin, the attribution is messy, and the churn is brutal. Use GEO as a tactic inside a larger product or content play, and you will sleep better—and so will your accountant.

Good luck, and may your next experiment last longer than fifteen days.