What does getting named actually mean?
A business can appear in an AI answer without earning a recommendation or a customer. Before commissioning GEO work, agree which result matters. We use GEO here to mean improving how a business can be discovered and represented in generated answers. It is a working description, not a separate licence or guaranteed placement.
| Outcome | What to record | What it does not prove |
|---|---|---|
| Mention | Your business name appears in the answer. | The answer approves of you. |
| Citation | A supporting link points to your page. | Your business is recommended. |
| Recommendation | The answer proposes you as a suitable option. | A buyer visited or contacted you. |
| Referral visit | An identifiable visit reaches your website. | The visitor is a good fit. |
| Qualified lead | An enquiry meets your agreed service and fit criteria. | It became a paid project. |
These are our reporting definitions. Record them separately rather than turning every appearance into one impressive score. An answer can cite your explanation of a service while recommending another supplier. That is useful visibility, but it should not be reported as a place on the buyer's shortlist.
For a Bangkok owner, the practical target might be accurate discovery for a particular service, language and area. For a European marketing lead, it might be being considered for a defined project type. Start there, rather than chasing every question containing your industry.
What should we fix first?
Google says ordinary SEO principles apply to AI Overviews and AI Mode. A supporting page must be indexed and eligible for a search snippet; meeting the requirements does not guarantee inclusion. Google does not require a special AI file or special structured data for these features. Google's AI search guidance.
Our order of work is to check access, then clarity, then evidence, then measurement. A supplier should be able to show the problem on the actual page before proposing a large content programme.
| If this is true | First action | Evidence to request |
|---|---|---|
| Your service page is not indexed. | Investigate search eligibility. | URL inspection and the specific access or indexing finding. |
| People cannot tell what you sell. | Rewrite one service page. | Scope, fit, location, proof and a clear next step. |
| Your name or hours differ across profiles. | Correct the business facts. | An approved record and updated public destinations. |
| You have visibility but poor enquiries. | Review the offer and enquiry journey. | Relevant visits, enquiry quality and rejection reasons. |
| Nobody has measured a baseline. | Run a small repeatable observation. | The questions, dates, answers and source links. |
Paying for a tracker first can be sensible when there is already a clear measurement owner. Otherwise it risks documenting problems that nobody has agreed to fix. A small spreadsheet is enough to establish what you need to learn.
Which business facts deserve attention?
Keep the official business name, actual service area, contact route and operating details consistent. Google's local guidance describes relevance, distance and prominence as the principal factors behind local results. Complete information and genuine reviews matter there. This is guidance for local Google results, not proof of a particular ChatGPT ranking formula. Google's local ranking explanation.
We would make each main service page answer five buyer questions: what is included, who it fits, where it is available, what evidence supports it, and what happens after an enquiry. Use examples you can substantiate. A project description should distinguish work you delivered from work you supported. Publish a useful price basis when possible, or explain the inputs needed for a quote.
Good evidence can be a dated process explanation, an attributed project, a clearly bounded comparison or an answer to a recurring customer question. Google asks creators to assess originality, experience, sourcing and whether readers leave with a useful answer. It does not prescribe a preferred word count. Google's helpful content guidance.
Bangkok and Europe need usable language journeys
Translate the journey customers actually use, including the service details and enquiry confirmation. Ask a fluent reviewer to check meaning, not just spelling. For multilingual sites, Google recommends separate URLs for language versions and language annotations. Google's multilingual guidance. Our recommendation is to test Thai and English questions separately, and the relevant European languages separately, because each set addresses a different audience.
Earn third-party coverage where your customers already make decisions: a relevant association, partner, publication or genuine review channel. Keep the description accurate. We would not buy unrelated directory entries simply to manufacture a larger mention count.
Does allowing AI bots give away our content?
Search access and model training are different decisions. OpenAI documents OAI-SearchBot for search and GPTBot for content that may be used in training. Their controls are independent. A site can allow the search crawler while disallowing the training crawler. ChatGPT-User handles some user-triggered visits and is not the control for search inclusion. OpenAI crawler documentation.
Google-Extended controls certain Gemini training and grounding uses. Google explicitly says it does not affect Google Search inclusion or ranking. Google's crawler documentation. Googlebot and search preview controls govern the relevant Google Search access. Google's AI search guidance.
Ask the website owner to check both the published crawler rules and the hosting security settings. Our recommendation is to agree the content policy before changing access. An invoice portal or private customer area should not become public to pursue citations. Public guides and private business records serve different purposes.
There is no verified basis in these sources for promising that an llms.txt file or knowledge bundle will earn a citation. If you maintain one for another useful purpose, keep it consistent with the website. Give it a maintenance owner and avoid presenting it as a ranking shortcut.
Use this measurement worksheet
Choose a manageable set of actual buying questions. Our suggested starter set covers discovering a supplier, comparing approaches, checking suitability and understanding cost. Include unbranded questions as well as your company name. Keep questions neutral: asking an assistant to explain why your company is best is not a discovery test.
Illustrative prompts include: find a provider for this service in Bangkok; what should a small business compare before buying this service; which suppliers suit this stated requirement; and what information is needed to quote this project. Replace the service and constraints with your own. These are proposed test inputs, not measured customer searches.
| Worksheet field | What we record |
|---|---|
| Question and intent | Exact wording, language, location supplied and buying stage. |
| Test conditions | Date, platform, visible mode, signed-in state and a fresh conversation. |
| Answer evidence | Saved answer or screenshot, cited URLs and whether an AI answer appeared. |
| Your result | Mention yes/no; citation URL; recommendation and its wording. |
| Accuracy | Wrong address, outdated service, unsupported claim or confused identity. |
| Business outcome | Separately observed visits, enquiries and qualified opportunities. |
| Change log | Page changed, reason, owner, date and next review. |
Repeat the same set under documented conditions. Keep missing answers and failed checks in the record. Report the count and denominator, such as mentions divided by completed observations, and explain exclusions. Do not relabel that sample as your share of all AI searches. We recommend saving raw answers so someone else can challenge the scoring.
Google includes AI feature traffic within the Web search type in Search Console. Those totals should not be presented as a clean AI-only channel. Google's performance guidance. Use available referral evidence for identifiable visits, and ask enquirers how they first heard about you. Keep self-reported attribution separate from measured clicks. Unknown should remain unknown.
How do we judge whether it is working?
We recommend evaluating visibility and commercial outcomes side by side. More accurate descriptions can be worthwhile even before enquiry volume changes. More mentions alongside fewer suitable enquiries may point to poor positioning. A recommendation for a service you no longer offer is a correction task, not a win.
Pros and limits of this approach
The advantage is that improving your service explanation, evidence and enquiry journey also helps people who reach you through referrals or ordinary search. The limitation is attribution. A small observation set cannot prove what every buyer sees, and one changed answer does not prove your page edit caused it.
If a monitoring supplier proposes a visibility score, ask which prompts, markets, languages and answer modes it covers. Ask whether the score counts mentions, links or recommendations, how failures are handled and whether you can export the evidence. A useful dashboard makes decisions easier; it does not make uncertain attribution certain.
Keep a change log and avoid altering the entire site between observations. Review a meaningful period for your sales cycle and compare enquiry quality. Our suggested experiment is one service page, one audience and one explicit hypothesis, such as whether clearer service boundaries reduce irrelevant enquiries.
Common mistakes and your next step
- Counting a citation as a recommendation or a sale.
- Testing only your own name, then claiming buyers are discovering you.
- Inventing statistics, testimonials or project results to look authoritative.
- Publishing near-identical location pages without distinct customer value.
- Assuming a training crawler controls search access.
- Changing the reported question set when results are disappointing.
Google's spam policies cover scaled content created primarily to manipulate rankings and doorway pages. Do not commission a batch of thin city pages simply because generation is easy. Google's spam policies.
At DLVX, our existing website planning guide separates search readiness from results and makes the enquiry journey part of acceptance. This guide adds a practical observation method. It does not claim a measured uplift for DLVX or our clients.
Next, pick the service you most want to sell. Check its public facts, save a baseline using the worksheet and correct the most consequential gap. Agree who will review the evidence again. The first useful deliverable is a better buying decision and a repeatable way to observe it.
Sources and review date
Reviewed 12 October 2026. Prices and plans are as of October 2026. DLVX recommendations are editorial judgements, not a vendor ranking or a promise of results. Recheck the linked terms before buying or connecting business data.
- Google: AI features and your websiteChecked 2026-10-12. Ordinary SEO requirements apply; indexed snippet eligibility required; no inclusion guarantee No special AI files or schema required AI feature traffic included in Search Console Web reporting Googlebot and preview controls govern Search access
- Google: improve local rankingChecked 2026-10-12. Local relevance, distance and prominence Complete information and reviews help local presence
- Google: helpful contentChecked 2026-10-12. Originality, experience and sources in content self-assessment No preferred word count
- Google: multilingual websitesChecked 2026-10-12. Separate URLs and language annotations for language versions
- OpenAI: crawler documentationChecked 2026-10-12. OAI-SearchBot search and GPTBot training controls independent ChatGPT-User not the search inclusion control
- Google: common crawlersChecked 2026-10-12. Google-Extended does not affect Search inclusion or ranking Google-Extended governs certain Gemini training and grounding uses
- Google: spam policiesChecked 2026-10-12. Scaled content abuse and doorway abuse policies
Want to apply this to your team? Get a free AI plan.
