You’ve been working on your SEO. Perhaps you’re paying an agency, managing it in-house or doing what you can between customer calls.
Now you’re hearing about AEO, GEO and LLMO. Another thing to learn. Possibly another thing to pay for.
Before spending more, understand how AI answers find information, what your existing SEO work covers and what still needs attention. Having an SEO plan doesn’t mean every website problem has been solved.
Your SEO work still counts
| Term | Meaning | What it focuses on |
|---|---|---|
| SEO | Search engine optimisation | Helping people discover your pages through search engines |
| AEO | Answer engine optimisation | Making information useful for direct answers, including snippets and answer boxes |
| GEO | Generative engine optimisation | Helping your business or information appear in AI-generated answers |
| LLMO | Large language model optimisation | Making content and business information usable in answers from language models; often overlaps with GEO |
These terms describe overlapping efforts to help people find your business through search results and AI answers. Agencies and software companies don’t always use them consistently. These labels overlap. They do not establish four separate sets of work.
SEO helps people find your pages through search engines. AEO usually refers to appearing in direct answers. GEO and LLMO commonly describe improving how businesses and information appear in answers from generative AI systems. People don’t always use these terms consistently.
Google says its established SEO practices remain relevant to its generative search features. Useful content, accurate business information and accessible pages still belong in the plan.
AI search adds questions about which sources are used, how your business is described and whether those answers help someone choose you. You can build on existing SEO work without starting from scratch.
Before another package, check the website
A customer shouldn’t have to wait through a slow page, dismiss several pop-ups and search for a booking button.
Search systems also need to reach and read your information. Check whether important pages are accessible, their content loads correctly and the website works on mobile.
Neil Patel’s technical SEO guide includes these foundations. Aleyda Solis’s AI-search checklist also checks whether systems can retrieve and read priority pages before recommending fixes.
An AI-built website can look finished while still needing technical checks. So can a website built without AI. Large images, unnecessary scripts and slow server responses can frustrate visitors.
Google can process JavaScript, the code behind many interactive website features. Your developer should still check that important content and links appear correctly. Google’s JavaScript guidance explains what to check.
Run a page through PageSpeed Insights and share the results with whoever manages your website. Google says Core Web Vitals contribute to its ranking systems, but perfect scores don’t guarantee top rankings. Fix problems that affect visitors instead of chasing 100.
You can investigate AI answers alongside these checks. Known access failures or broken booking steps deserve attention before expanding your monitoring budget.
How an AI answer gets its information
An AI answer combines three inputs:
- Learned during training. The knowledge cutoff varies by model.
- Your question and conversation.
- Web search, when used. Web search can bring in newer information. Reading a page does not permanently update training.
An AI assistant can draw on information learned during training and information available in the conversation. When web-search tools are available and used, it can also look things up online.
During training, a model processes examples such as books, webpages, articles and code. It learns patterns stored as numbers, rather than keeping a neat library of pages it can reopen.
A large language model, or LLM, uses those patterns and the information available to it to generate an answer. Its knowledge cutoff indicates roughly how recent its training information may be; it doesn’t mean the model knows everything published before that date.
Updating your website doesn’t immediately update what the model learned during training.
When an AI assistant uses web-search tools, it can look up information online and use the results while answering. It might retrieve your current opening hours instead of relying on older information. Reading a page during that lookup doesn’t necessarily mean it permanently learns it.
Google, ChatGPT and Claude can look things up differently
| Platform | How it works |
|---|---|
| Google AI search | Finds relevant pages in Google’s search index. It can run several related searches, then use retrieved information in its answer. |
| OpenAI web-search tools | Can look up information before answering. Reasoning models can examine results and decide to search again. |
| Claude web-search tools | Claude can request searches, receive results and repeat the process before answering with citations. |
Check more than one platform. Your business may appear in one AI answer and be absent from another. Search capabilities vary by product, model and settings.
These descriptions come from Google, OpenAI and Anthropic. The latter two sources describe developer tools, not every decision inside consumer ChatGPT or Claude.
Access and visibility can differ between platforms. Appearing in one doesn’t establish that you appear in all of them.
Check whether the AI assistant used web search and supplied sources. A confident answer alone doesn’t establish either.
Finding a page is only the beginning
A simplified web-assisted process, from your website to an AI answer:
- Page accessible
- Found through web search
- Relevant information selected
- Used in the answer
- Possibly cited
Being found does not guarantee a citation or recommendation.
When checking an answer, look at whether it uses your information, links to your page or recommends your business.
A page can be available without being selected for a particular answer. An AI assistant might use your information without recommending your business, or recommend your business using another website.
A link also doesn’t tell you whether someone clicked it, contacted you or bought something.
Mentions, citations and recommendations mean different things
| Result | Example answer |
|---|---|
| Mention | Willow Dental is a dental practice in London. |
| Citation | Treatment suitability depends on your dental health. Source: Willow Dental guide |
| Recommendation | Willow Dental may be worth considering because it offers the treatment you need. |
Record mentions, citations and recommendations separately so you know what each result represents. Fictional business and illustrative answers.
A brand mention names your business. A citation links to a source used in an answer. A recommendation suggests your business as an option for someone’s needs.
An AI assistant could cite your guide to dental implants without recommending your clinic. It could recommend your clinic while citing a directory or an independent article.
Businesses have long sought relevant third-party coverage through digital PR, reviews, partnerships and other off-page SEO work. Look for opportunities to contribute expertise, a case study or research. HARO, operated by Featured, can help you find journalists seeking sources.
Respond where your experience fits, with examples you can support. Google advises against seeking inauthentic mentions, so avoid manufacturing coverage just to increase a count. Publication, links and customer enquiries aren’t guaranteed.
What is being sold under AEO, GEO and LLMO?
| Area | What you may see advertised | What the work involves | How it relates to existing SEO |
|---|---|---|---|
| Website foundations | Crawler access and website readiness | Checking whether crawlers can reach pages and read their content | Established technical SEO, with access checks for the AI-search platforms you want to reach |
| Website foundations | Schema implementation | Adding code that labels products, addresses and authors | Established technical SEO; Google requires no special AI schema |
| Research and content | Customer-question and competitor research | Finding relevant customer questions and checking which businesses and sources appear in AI answers | Builds on search-intent and competitor research, with additional AI-answer checks |
| Research and content | Answer-formatted content | Organising pages so readers can find clear answers | Overlaps with useful content and readable structure; no fixed answer length is required for Google’s AI features |
| Research and content | Content gaps and supporting evidence | Adding missing explanations, case studies, original data or expert contributions | Established content work; may require new research or subject expertise |
| Business identity and authority | Entity engineering | Making your business identity, services and profiles clear and consistent | Overlaps with business listings, structured data and accurate company information |
| Business identity and authority | Citation seeding/brand mentions | Seeking relevant third-party coverage about your business | Overlaps with digital PR and off-page SEO; placement quality and method matter |
| Measurement | Prompt panels / AI visibility tracking | Repeating selected questions across AI platforms and recording answers | Additional measurement that may require tools or staff time; a sample of answers, not total customer exposure |
| Measurement | Citation-source monitoring | Checking cited pages for relevant, inaccurate or outdated information | Adds AI-source research to monitoring; corrections may need separate work |
Compare these tasks with work already completed or planned. An audit recommendation, an implemented fix and ongoing monitoring are different deliverables.
Your current SEO work may already cover some of these tasks. Check what has been done, not only what appears in the service description.
An audit might recommend a fix without anyone implementing it. A monitoring report might flag outdated information without anyone correcting it. Ask whoever handles the work what they changed, what remains unresolved and who will handle it.
Further research, content or development may still be needed, even when the tasks fall under familiar SEO work.
Structured data is code that labels information on a page, such as a product, business address or article author. Google says you don’t need special schema or AI-specific files to appear in its AI Overviews or AI Mode. Relevant structured data can still support wider SEO.
Clear answers help readers. Google’s AI features don’t require every paragraph to follow a fixed word count.
You can check a public page with Google’s Rich Results Test or the Schema Markup Validator, without backend access. These check markup, not whether every business detail is accurate or whether AI will cite you.
What about robots.txt and llms.txt?
robots.txt gives crawlers instructions about which pages they may access.
Your website doesn’t need this file to be crawled. When it’s missing and returns a normal “not found” response, Google assumes there are no crawl restrictions. Existing rules can accidentally block pages you want discovered.
llms.txt is a proposed guide to a website’s information for AI tools that use it. Think of it as a reading list, rather than permission to enter.
Google says it ignores llms.txt for Search, including its AI features. You don’t need Markdown copies of your pages either.
These formats can help compatible AI tools navigate information, particularly software documentation. That doesn’t establish that adding them will improve your business’s visibility in AI answers. Google’s position also isn’t a rule for every other platform.
Letting an AI search tool find your pages differs from allowing collection for model training. OpenAI and Anthropic document separate crawlers for these purposes.
Ask whoever manages your website whether existing rules or security settings block the search crawlers you want to reach. Don’t remove restrictions without understanding why they were added.
If you’d like help with these checks, MBAgency can walk through them with you, including how to keep search access separate from AI-training permissions.
Try a small check without paid monitoring software
Choose a service or product that matters to your business and a question customers ask when deciding whether to buy.
You, your team or your agency can use this checklist to investigate selected AI answers. It can uncover missing or incorrect information; it won’t establish overall SEO health or measure every potential customer’s exposure.
- Choose a customer question. Write down three questions from enquiries, sales calls or support. Choose one about a service or product that matters to your business.
- Ask an AI assistant. Include relevant location or requirements. Check that web search was used and sources were supplied.
- Inspect the sources. Open the cited pages. Check their accuracy and whether the answer represents them fairly.
- Review your page. Does it explain your service, suitability, limitations and how to contact you, book or buy?
- Choose an action. Correct details you control, request an external correction or investigate missing information. Record uncertainty rather than guessing.
- Record and repeat. Save the date, platform, question, answer, sources and finding. Add the action and person responsible. Repeat later; answers vary.
Start with one question about a service or product that matters to your business. Record what you find, including anything you cannot yet explain.
AnswerThePublic can suggest related questions. Use your own enquiries and customer conversations to choose which ones to investigate first.
What this check could look like
This is a hypothetical example, not a real client or a real test result.
Imagine a clinic in Johor Bahru that offers acne scar consultations. A customer might ask an AI assistant:
Which clinics offer acne scar consultations in Iskandar Puteri, Johor Bahru? Please search the web and include clinic locations and sources.
The question doesn’t name the clinic. You might run it across a few ChatGPT accounts, again in fresh temporary chats with personalisation turned off, and once in Claude.
The results could differ. The clinic might appear in some answers and not in others. One answer might cite the clinic’s acne-scar page, while another cites mostly directory listings and a third-party page about a different clinic. The other clinics listed could also change from chat to chat.
Even then, it would be hard to tell why the clinic appeared in one answer and was absent from another. A single check can’t show whether AI answers lead to enquiries either. MBAgency can repeat checks like this over time and, with website and booking tracking in place, follow identifiable AI referrals through to enquiries and consultations.
Creating and maintaining useful service and location pages is established SEO work. If you already pay for that, check whether the proposed AEO or GEO service would improve those pages, add something missing or repeat work already completed.
What if your business doesn’t appear?
Check what you asked. A question about how a treatment works may produce an explanation rather than a list of clinics. Asking where to book that treatment in a particular city invites a different answer.
If other businesses are recommended, open the cited pages. Look for information about services, location, suitability and experience, then compare it with your own pages and profiles.
You may find useful details your pages are missing. Those sources won’t tell you exactly why the AI recommended one business over another.
Rand Fishkin‘s research on repeated AI recommendations found that the businesses listed and their order could change when the same question was repeated. Some brands still appeared frequently. Repeat your question in fresh chats before treating one missing mention as a problem.
You can also try the free Ahrefs AI visibility checker or Semrush AI visibility checker. These check samples of AI answers. They can’t tell you what appeared in everyone’s private conversations.
Try the tools you already have
Monitoring software can save time when you repeat checks across many questions, platforms or locations.
Ahrefs Lite and higher plans include a limited custom-prompt tracking allowance. Starter doesn’t. One check means one prompt run on one platform in one location. Each repeat uses more of the allowance.
If you use Semrush, check which AI-visibility features your subscription provides. Semrush One combines its SEO and AI Visibility toolkits. Access in other subscriptions differs.
Before buying another tool, try the relevant features you have. Decide who will review the results and make any necessary changes. An agency such as MBAgency can handle that work when your team lacks time or expertise.
Check what each number means. Ahrefs’ Brand Radar impressions are estimates calculated by its model. They aren’t a count of people who saw your business in an AI answer.
The report looks great. How many bookings did we get?
Being visible doesn’t guarantee leads or sales
An AI answer may give someone enough information that they don’t visit the websites it cites.
In Pew’s study of US Google browsing in March 2025, people clicked a traditional result during 8% of visits with an AI summary, compared with 15% without one. They clicked a link inside the summary during 1% of visits containing a summary.
These are rates for visits to results pages, not individual listings. The study didn’t establish separate rates for people looking to buy, so these averages aren’t a forecast for your business.
Look at what your content helps someone decide. Educational guides can attract people still researching. Comparisons, case studies and information about costs, suitability and what happens next help them evaluate your business.
If your website has plenty of general guides but little help choosing or booking, review those gaps before publishing another guide.
Aleyda Solís separates AI presence, website readiness and business impact. You need to check both whether your business appears and whether it gains suitable customers.
For a service business, record enquiries and whether they fit your service, location and budget. For an online store, track completed purchases, revenue and profit where you can.
Google Search Console’s generative AI performance report shows impressions in supported Google AI features. It doesn’t separately measure clicks or conversions. Your site may not have enough impressions to show the report.
Website analytics can identify some visits from AI platforms. Check whether those visitors submit an enquiry or complete a purchase. Ask new customers how they found you. If you don’t know the source, record it as unknown.
What now?
Use the comparison and checklist to choose what needs attention for a service or product that matters to your business.
| What you’ve found | Where to start |
|---|---|
| A known access problem or broken booking/purchase step | Prioritise investigating and fixing it. |
| Incorrect information or a missing customer answer | Correct information you control; request external corrections where possible. |
| Poor-fit enquiries | Review the offer, audience and expectations set by the relevant pages. |
| No appearance, with no clear cause | Gather more observations before choosing a fix or buying additional work. |
Start with a problem you can identify. One missing AI mention doesn’t tell you what needs fixing.
You can do the checks yourself, ask your team or have an agency such as MBAgency handle them. You don’t have to become an SEO specialist before getting help.
If you already work with an agency, ask how it addresses AI search within your current service. Before agreeing to additional work, ask what it will change, why it’s needed and how results will be reviewed.
If you’re unsure where to focus, send us your website and what you’re trying to improve. We’ll set up a short call to understand your situation and suggest a few practical steps, whether or not you end up working with us.
Book a call with MBAgency