In August 2026, Taiwan saw at least seven announcements of GEO (generative engine optimisation) service launches, all inside three weeks.
The range was wide: a payments provider moved in, a manufacturing consultancy planted a flag in Taichung, a retail commerce platform announced a strategic partnership, and a dental-specific AI search service launched locally. Add infrastructure players treating AI crawler traffic as a new battleground, and packages bundling AI content tooling with digital PR.
The market has formed. Which means proposals are coming your way — and they will use remarkably similar language.
This article compares no vendors and names none as better or worse. It offers something else: a method for testing proposals against official documentation — because the biggest problem in GEO right now is not which vendor is strongest. It is that a lot of claims have no basis at all.
💡 Got a GEO proposal and unsure how to judge it? We will read it through with you, free. 👉 Free consultation on LINE
Key takeaway: Judging a GEO proposal is less about whether the vendor is technically strong and more about whether the claims in it match the official documentation. Google states plainly that appearing in AI features requires "no additional requirements... nor other special optimizations." Any proposal implying proprietary technique, or asking you to build AI-specific files first, deserves hard follow-up.
Start With a Fact That May Surprise You
We sell GEO services ourselves, and this still needs saying —
Google's own documentation answers the question "how do I optimise to appear in AI Overviews and AI Mode" like this:
"There are no additional requirements to appear in AI Overviews or AI Mode, nor other special optimizations necessary."
"You don't need to create new machine readable files, AI text files, or markup to appear in these features."
That is unfavourable to the entire GEO industry, ours included. But it is the official position, and you should know it before evaluating any proposal.
So what does a GEO service actually do? The honest answer: the same old things — content quality, technical crawlability, information structure — with a different order of priority and a different way of verifying results. AI search expands one question into many sub-questions before looking for answers, so "does your content cover those sub-questions" matters more than "are you on page one."
The real difference is in measurement and trade-offs, not in secret technique.
Which gives you your first test: if a proposal implies techniques nobody else has, or frames GEO as something entirely separate from SEO, that does not match what Google says.
Six Questions to Ask
Run these six against any proposal. Whatever cannot be answered is your risk register.
1. How do you define success, and what number verifies it?
The most important question, because AI search measurement has no industry standard yet.
Tools define "AI visibility" very differently: who chose the query set, which engines are covered, whether a mention and a citation count as the same event, how the score is computed. Two tools measuring the same brand will not produce comparable scores — not because one is wrong, but because they are measuring different things.
Follow up with: whose tool produced this score? Who chose the queries, and how many? Which engines? If we switched tools, would the number change?
If they cannot answer, that score cannot be your acceptance criterion.
2. Are you promising citation, or recommendation? They are not the same
A study of purchase-decision queries (Shero Commerce) analysed 1,851 sources cited across Google AI Mode, ChatGPT and Perplexity:
- Only 2.8% were brand-owned pages
- The largest source category accounted for 59% — third-party review sites like Good Housekeeping and Verywell Fit
In other words, in a commerce context, when AI recommends your brand it usually cites someone else's site.
For a proposal, that means: promising "your site will be cited by AI" is harder than it sounds in some contexts; and if the goal is "AI recommends your brand," the work may include earning presence on third-party review sites, which is a different job entirely.
Follow up with: are you working to get my site cited, or my brand mentioned? Do those goals use the same method?
(Scope note: that study covers purchase-decision queries in a commerce context and does not generalise to all query types.)
3. How long until something changes, and what do I watch meanwhile?
AI search results shift on their own rhythm, not identical to classic rankings. A sound proposal tells you what to watch in the first three months, when numbers become meaningful, and how you would know if it is not working.
Follow up with: if three months pass with no change, how do we tell "not yet" from "wrong direction"?
A proposal with no stop-loss condition usually has not thought through its success condition either.
4. Who writes the content, and how much human time goes in?
Google's spam policy targets content produced at scale to manipulate rankings with no value to users, and states explicitly that this applies "no matter how it's created."
So the question is not "do you use AI" — using AI is not a violation. The question is how many editing minutes go into each piece, and whether the content contains anything only your company could supply: real cases, your own data, the actual questions your customers ask.
Follow up with: from generation to delivery, how long does a human spend per article? What material do I need to provide?
If the answer is "nothing, we handle everything," be careful — without your material, what comes out is content anyone could generate.
5. Who owns the output?
Including: copyright in the articles, the files after they go live, the tracking accounts and reports, and what you keep when the contract ends.
Follow up with: after the contract, is this content still mine? Do the reporting accounts transfer to me?
6. What is and is not in the price?
Common grey areas: article count and length, images, publishing, internal link changes, technical modifications (who edits the site code), and monthly analysis reporting.
Follow up with: how many pieces are included? How is overage charged? Do you make the technical changes, or does my engineer?
Four Claims That Need Follow-Up
Not that these are necessarily dishonest — just that hearing them should prompt one more question.
"We can get you onto the AI recommendation list"
AI answers are generated at request time; the same question can produce different answers at different moments and for different users. There is no list to be placed on.
Follow up: what metric, under what conditions, are you actually describing?
"First we need to build your llms.txt / AI-specific file"
Google states plainly that these files are not needed. Ahrefs analysed 137,000 websites and found that among sites with an llms.txt, 97% received not a single request to it across an entire month — not merely unread by AI, but never fetched at all.
Follow up: which AI provider commits in official documentation to reading it?
(This does not make it forbidden — it is cheap, and the exercise of summarising your own site has some value. But if it is the centrepiece of the proposal, that is a problem.)
"We have proprietary GEO technology"
Set against Google's "nor other special optimizations necessary." Real differences come from topic judgement, content depth, and measurement method — not technique.
Follow up: what does this technology concretely do, and how does it differ from ordinary content and technical SEO?
"We guarantee results within N months"
Nobody can guarantee the behaviour of AI engines. What can be committed is output (how many pieces, which pages changed, what the monthly report contains), not outcome.
Follow up: does that guarantee go into the contract? What is the remedy clause if it is missed?
How to Define an Outcome You Can Verify
Rather than accepting a vague visibility score, split acceptance into two layers.
Layer one: output (fully within the vendor's control — put it in the contract)
- How many pieces, how long, covering which topics
- Which existing pages get revised, and what changes
- Which fields the monthly report contains
Layer two: outcome (neither side fully controls it — use it to check direction)
- Referral traffic from AI tools in your own GA4
- Impression movement for your target query group in Search Console
- A fixed question set, on fixed engines, at a fixed cadence, recording whether your brand comes up
Layer two is not about the absolute number. It is about measuring with the same ruler each time. Recording it yourself is more reliable — and cheaper — than buying a score whose computation you cannot inspect.
FAQ
Seven launches in a month — is this a bubble?
Not necessarily. AI search has genuinely changed traffic structure, so the demand is real. But real demand and supply quality are separate questions. When a market forms fast, claims outrun track records — which is exactly why a method for testing proposals matters.
If Google says no special optimisation is needed, is a GEO service necessary at all?
Yes, provided you understand what it does. Google's point is that there is no proprietary technical requirement, not that content quality and structure are irrelevant. The value of GEO work is in topic selection, content depth, information structure and measurement — all of which always mattered; the priority order changed.
Can I do it myself?
At small scale, yes. Start with three things: confirm AI crawlers can reach your site; turn your most important questions into content that genuinely answers them; and record monthly, with a fixed question set, whether your brand gets mentioned. Then decide whether to outsource.
Can I trust the visibility score in a proposal?
As a reference, yes; for cross-vendor comparison, no. Query sets, engine coverage and scoring all differ. To use it as an acceptance criterion, the definition has to be written into the contract.
Should I hire a GEO specialist or my existing SEO vendor?
More important than specialisation is whether they can explain what they do and how it gets verified. GEO and SEO work overlap heavily, and an SEO team that explains itself clearly is usually safer than a GEO team that speaks in mystique.
Conclusion: When the Market Is Noisy, Official Documentation Is the Only Anchor
Seven entrants in three weeks means you will hear many versions of the same pitch. They will use similar vocabulary, quote different numbers, and promise things that sound alike.
Telling them apart is not about who sounds more expert. It is about taking each claim in the proposal and checking it against official documentation. What matches, keep discussing. What does not, follow up until it does.
And if a proposal volunteers something inconvenient to itself — like "Google says no special optimisation is needed" — that is usually a good sign.
💡 Want a second pair of eyes on a GEO proposal? We will run the six questions with you. No fee, no pitch. 👉 Contact AI SEO Hacker
Further Reading
- How to Measure GEO Performance: AI Search Visibility Metrics
- What Is llms.txt? Should You Set One Up? The Official Position
- How to Choose a Marketing Agency
- Will Google Penalise AI-Written Content?
- The Complete Guide to Content Marketing Outsourcing
References
- AI features and your website — Google Search Central
- Spam policies for Google web search — Google Search Central
- ECPay and Spider Network launch GEO services for merchants — TechNews (7 Aug 2026)
- Reed Tech plants a flag in Taichung, targeting GEO for manufacturers — PChome Online News (6 Aug 2026)
- Shengeng Marketing announces its "GEO core triangle" — Leadfin (10 Aug 2026)
- Retail GEO insights: strategic partnership with SHOPLINE — ENN Taiwan (23 Aug 2026)
- IDAEO launches in Taiwan targeting dental search — CNEWS (18 Aug 2026)
- SkyCloud launches AICDN — iThome (4 Aug 2026)
