This DIY GEO and AEO guide, as of October 2026, gives you seven steps that get your business named correctly in AI answers, upstream of your leads and sales. You list the buyer questions, write one verified answer to each and state each fact in a plain sentence. You open your site to the AI crawlers, mark up each page, earn independent confirmation and test the answers in ChatGPT, Claude, Gemini and Perplexity. A writer, a developer and one named owner of the monthly check do all seven without an agency. You hire help at five points, each with a named sign.
Written by Kevin Dam, founder of Aemorph, who has practised SEO since 2010 and sits on the Forbes Agency Council. Aemorph is a Google Partner. LinkedIn
DIY GEO and AEO Covers Four Things Your Team Controls
DIY GEO and AEO, as of October 2026, puts your visibility in AI answers in your own team's hands through four things you control: your facts, your pages, your crawler settings and your checks. GEO, Generative Engine Optimisation, is citation inside the AI chats. AEO, Answer Engine Optimisation, covers Google AI Overviews, featured snippets, answer boxes and voice answers. Both build on SEO, search engine optimisation, which earns your place in Google's regular results. Each AI engine selects the sources an answer cites, outside the control of any business, so a DIY plan works on the inputs and tests the outputs.
| Surface | Discipline | What you control | What you do not control | First DIY check |
|---|---|---|---|---|
| ChatGPT, Claude, Gemini and Perplexity | GEO | The facts your own pages state, your crawler-access settings, and the descriptions on profiles you manage | Which sources an engine retrieves and which business an answer names | Ask the buyer's question in each engine and log the answer |
| Google AI Overviews | AEO | The page's technical accessibility, text-form content, and indexing and snippet settings | Whether Google shows an overview for the query, and which pages it links | Search the question in Google and note the supporting links |
| Featured snippets and answer boxes | AEO | A direct answer near the top of the page that owns the question | Whether Google's systems pick the page for the box | Search the question and note which page holds the box |
| Voice answers | AEO | One short, plain answer per question | Which single answer the assistant reads back | Ask the question aloud on a voice assistant |
What a DIY plan controls on each AI answer surface.
GEO Leads a DIY Plan and AEO Follows
GEO leads a DIY plan because the chat assistants are where a buyer asks for a shortlist and reads a written answer that presents recommendations. Source links and citation displays vary by engine and mode. The same verified facts feed both disciplines, so the AEO work on Google's answer surfaces reuses the record the GEO work builds; split the two into separate projects and you do the fact-finding twice. Google's AI Overviews, featured snippets and voice answers draw on pages that state one clear answer, so the record your DIY pass builds is also the input that Aemorph's Answer Engine Optimisation (AEO) service optimises for Google's answer surfaces.
Three Things Sit Outside a DIY Plan
Three things sit outside a DIY plan: the engines' own choice of sources, public records that other people edit, and any guarantee on the result. You edit your own pages freely. Records such as Wikidata and the Google Knowledge Graph are edited and reviewed by their own communities and systems, so an edit there does not always stay in place. No tool and no checklist turns a DIY pass into a guaranteed citation.
The Seven Steps a Business Does Itself, With the Check for Each
You do GEO and AEO yourself, as of October 2026, through seven steps that run in order, from your buyers' questions to a dated test of the answers that shows whether AI names your business, and each step ends in a check your team runs on its own. AI keeps a map of what every business is and what it does. Each engine answers from what it has learned and retrieved about a business across many sources, and your website is one of them. Page volume alone does not make you ready for citation; one clear, verified answer per buyer question does. The sentence-level rule behind step 3 is set out where Aemorph writes each business fact the way a machine reads it.
| Step | What you do | Who does it | Check to run | Pass mark | Where the method comes from |
|---|---|---|---|---|---|
| 1. List the buyer questions | List every fact a buyer wants confirmed: services, problems solved, audiences, credentials, locations, delivery formats, and the showstopper questions on price, trust, fit, risk and timing | The owner or marketing lead, with a subject expert | Count the questions that have one clear answer on a page or record | Every question on the list has an answer, or a named gap | Aemorph's full-coverage attribute mapping |
| 2. Write one verified answer per question | A subject expert answers each question; one named person signs the answers off; wrong information on the website is fixed first | A subject expert and the sign-off owner | Read each answer against its source document | No answer contradicts another page | Aemorph's Sprint 1 practice: the point of contact signs off before publishing |
| 3. State each fact in one plain sentence | Put the subject, the verb and the specific detail next to each other | The writer | Read the sentence alone, out of its paragraph | The sentence still states who, what and the detail | Aemorph's machine-readable fact sentences |
| 4. Open the site to the AI crawlers | Set robots.txt rules for the search crawlers of each engine and keep key content in text form | The developer | Open /robots.txt and read the rule for each crawler | No search crawler is blocked by accident | OpenAI, Anthropic, Perplexity and Google documentation, October 2026 |
| 5. Mark up what the page already says | Add structured data that states the same values as the visible text | The developer | Run the full markup through the Schema Markup Validator and the Google-supported features through Google's Rich Results Test, then compare each value with the visible text by hand | Both tools report no errors and each value matches the visible sentence | Google Search Central, updated 10 December 2025; Schema.org's Schema Markup Validator |
| 6. Earn independent confirmation | Align the name, category and services on profiles, directories, industry bodies and press mentions | The owner or the PR lead | Compare each outside description with the website | One description everywhere a buyer or an engine looks | Aemorph's Resonance Network (digital PR and third-party citations); Chen and colleagues, September 2025 preprint |
| 7. Test the answers and log them | Ask each question in each of the four AI chats and keep a dated record | The named owner of the check | Log each question, each engine, the date and the answer | The answer names the business and states its facts correctly | Aemorph's cluster-level measurement |
The seven DIY steps and the check for each.
The Record: Buyer Questions, Verified Answers and Fact Sentences
Steps 1 to 3 build your verified record, the single approved set of facts about your business, and the record is where most DIY effort belongs, because every later step repeats what it says. A fact has three parts: the business, a thing that is true of it, and the specific detail. Aemorph lists those facts across ten areas of a business, from services to the surfaces where a buyer finds it, and the same list works for your team: Aemorph's attribute mapping methodology maps every fact a buyer wants confirmed.
A DIY record passes three tests. It:
- Answers each buyer question once, on the page where a buyer looks for it.
- Agrees with itself, so no two pages state the same fact in different terms.
- Names a person who signs off each answer before it is published.
A showstopper question is a buyer objection question that stops a purchase when an AI assistant answers it badly. Those questions go first on your list, because a wrong answer there puts a sale at risk.
Access and Markup: Crawlers, Text and Structured Data
Steps 4 and 5 make your record readable, and each of the four engine operators publishes the crawler names your site allows or blocks in robots.txt. OpenAI documents separate settings for search and for training: OpenAI's crawler documentation lists OAI-SearchBot, GPTBot and ChatGPT-User, and your site allows OAI-SearchBot to appear in ChatGPT search while it disallows GPTBot. OpenAI states that a robots.txt change takes about 24 hours to reach its search systems.
For Google's answer surfaces the requirement is ordinary search eligibility. Google states that a page shown as a supporting link in AI Overviews or AI Mode is indexed and eligible for a snippet, and that those features carry no additional technical requirements, in Google's guidance on AI features and websites (Google Search Central, updated 10 December 2025). The same page lists text-form content and structured data that matches the visible text among its basics.
| Crawler or indexing control | Operator | What it does | Effect of blocking it | Source and date |
|---|---|---|---|---|
| OAI-SearchBot | OpenAI | Surfaces websites in ChatGPT's search features | Blocking OAI-SearchBot excludes the site from ChatGPT search answers; navigational links are not excluded by this setting | OpenAI crawler documentation, October 2026 |
| GPTBot | OpenAI | Crawls content used in training OpenAI's generative AI foundation models | The content stays out of that training crawl; the setting is independent of OAI-SearchBot | OpenAI crawler documentation, October 2026 |
| Claude-SearchBot | Anthropic | Navigates the web to improve search result quality for Claude users | Anthropic's system does not index the content for search | Anthropic support article, updated 7 April 2026 |
| Claude-User | Anthropic | Visits a page when a Claude user asks a question that needs it | Anthropic's system does not retrieve the content in response to a user query | Anthropic support article, updated 7 April 2026 |
| ClaudeBot | Anthropic | Collects web content for Anthropic's generative AI models | Anthropic treats the block as a signal to exclude the site's future material | Anthropic support article, updated 7 April 2026 |
| PerplexityBot | Perplexity | Surfaces and links websites in Perplexity's search results; Perplexity states it is not used for foundation models | The site is not surfaced that way; Perplexity recommends allowing it | Perplexity crawler documentation, October 2026 |
| Google Search indexing | Supplies the pages AI Overviews and AI Mode link to | A page outside the index is not eligible as a supporting link | Google Search Central, updated 10 December 2025 | |
| Google-Extended | A robots.txt control token, not a separate crawler: controls use of content for Gemini model training and for grounding in Gemini Apps | No effect on inclusion or ranking in Google Search, by Google's own statement | Google crawler documentation, October 2026 |
AI crawlers and indexing controls a DIY team checks in robots.txt. Sources: OpenAI, Anthropic, Perplexity and Google crawler documentation, October 2026.
Crawler access makes a page eligible; it does not make an engine cite it. Schema.org markup on webpages is infrastructure: Google documents its use for rich results in Search, and OpenAI, Perplexity and Anthropic have not confirmed that their assistants use Schema.org markup on webpages.
Corroboration and Evidence: What Published Research Says Earns a Citation
Steps 6 and 7 rest on published research, and two papers give your team outside evidence on what earns a citation. The first tested content changes on a benchmark of 10,000 queries and reported that its methods lifted visibility in generative engine responses by up to 40%, and by up to 37% on Perplexity: the GEO study by Aggarwal and colleagues, KDD 2024. The second, a preprint of September 2025, found that AI search shows "a systematic and overwhelming bias towards Earned media (third-party, authoritative sources) over Brand-owned and Social content" (the preprint by Chen and colleagues, arXiv 2509.08919).
Both report findings under test conditions, across many queries rather than for one business. They support two DIY priorities: write claims a reader checks against a named source, and get the same facts confirmed on sites you do not own.
| Figure | Value | What it measures | Date | Source |
|---|---|---|---|---|
| Benchmark size | 10,000 queries | The query set used to test GEO methods | Paper revised 28 June 2024 | Aggarwal and colleagues, KDD 2024, DOI 10.48550/arXiv.2311.09735 |
| Visibility lift | Up to 40% | Change in visibility in generative engine responses on the benchmark | Paper revised 28 June 2024 | Aggarwal and colleagues, KDD 2024 |
| Visibility lift on Perplexity | Up to 37% | The same methods tested on a live generative engine | Paper revised 28 June 2024 | Aggarwal and colleagues, KDD 2024 |
| Robots.txt adjustment time | About 24 hours | Time for OpenAI's search systems to reflect a robots.txt change | October 2026 | OpenAI crawler documentation |
| Additional technical requirements for AI Overviews | None | Technical requirements beyond being indexed and eligible for a snippet | Updated 10 December 2025 | Google Search Central |
Sources: Aggarwal and colleagues, KDD 2024; OpenAI crawler documentation; Google Search Central.
The Checks to Run: From One Prompt to a Cluster Across Four Engines
The checks to run on DIY GEO and AEO, as of October 2026, climb three levels, from the site's access and markup to a prompt test of each buyer question to a cluster report across the four AI chats plus Google's answer surfaces, and only the top two show whether AI names your business. A team that stops at the first level knows its pages pass the access checks and nothing about what the engines say. The four chat engines are the surfaces Aemorph's Generative Engine Optimisation service builds citation on, and the same three levels apply to your team working alone.
| Level | What you test | Unit | What it shows | What it misses |
|---|---|---|---|---|
| 1. Access and markup | Robots.txt rules, indexing, text-form content and structured data | One page | Whether the page passes the initial access, indexing and text-format checks, and what a live request returns | What any engine says about the business |
| 2. Prompt test | Each buyer question in each of the four chat engines | One question in one engine | Whether the answer names the business and states its facts correctly | Other wordings of the question, and variation between sessions |
| 3. Cluster report | The share of a defined set of related questions that name the business | One cluster of buyer questions | Coverage across a buying decision, engine by engine, over time | Questions outside the defined set |
Three levels of DIY checking.
Level One: The Access and Markup Checks
The level-one checks use files and tools you already own, and they come before any prompt test. Run them in this order:
- Open /robots.txt and read the rule for each search crawler; Anthropic states that its bots honour robots.txt, in Anthropic's article on its web crawlers, updated 7 April 2026.
- Wait about a day before testing a robots.txt change: OpenAI gives about 24 hours for its systems to adjust.
- Confirm that each key page is indexed in Google and shows a snippet.
- Read each key page with scripts switched off, to see whether the answer survives as plain text.
- Validate the structured data and compare its values with the visible sentence.
- Request each key page as a live visitor and check the response status, any block from a security layer such as a firewall or content delivery network, and the answer text the page returns.
Level Two: The Prompt Test in Four Engines
The prompt test is the DIY check that matters most, because it shows whether the AI chats name your business: ask each buyer question in each of the four, and log what each one says. Aemorph's own test prompts follow four patterns, given as examples: "Best [service] company in Singapore", "Top providers for [category]", "Compare [brand] vs [competitor]" and "Who specialises in [specific problem]". For each answer, your log records the date, the engine and whether the session was logged out, and it notes whether the answer:
- Names your business anywhere in the answer, which counts as a mention.
- Recommends your business as an option to act on, which counts for more than a passing mention.
- States the facts as your website states them, with nothing wrong, outdated or missing.
Aemorph's own record shows what a dated entry looks like: in a logged-out ChatGPT session on 4 August 2026, ChatGPT cited Aemorph 3 to 4 times and named it as a source across four queries, "PSG SEO Singapore" among them. The paid version of this test is the AI Hallucination Diagnostic, which runs 50 to 150 prompts built on a business's showstopper questions across the same four engines.
Level Three: The Cluster Report and the Search Data
A cluster report turns your prompt log into one number per buying decision: the share of tested answers in a prompt cluster, a defined set of related buyer questions, that name your business. An illustrative report line reads "named in 7 of 10 tracked questions, up from 4 of 10". Aemorph reports the same way for clients, and Aemorph's guide to cluster-level measurement for AEO and GEO sets out the four steps.
Two free search-engine reports add data a prompt log lacks. Bing Webmaster Tools added an AI Performance report in public preview on 10 February 2026, covering Microsoft Copilot and AI-generated summaries in Bing, with total citations, average cited pages, grounding queries, page-level citation activity and visibility trends over time: Microsoft's announcement of AI Performance in Bing Webmaster Tools. Google counts traffic from its AI features inside the Search Console Performance report, under the Web search type (Google Search Central, updated 10 December 2025).
Click counts understate AI visibility. In Pew Research's March 2025 study of 900 US adults, users clicked a traditional result in 8% of visits to a search page with an AI summary, against 15% without one, and clicked a source cited inside the summary in 1% of visits: Pew Research Center's analysis of clicks on search pages with an AI summary, 22 July 2025.
When DIY Is Enough, and the Five Points Where It Stops
DIY GEO and AEO is enough, as of October 2026, for a team with a writer, a developer who ships fixes and one named owner of the monthly check, so you pay for help only where your own log shows a gap; it stops being enough at five points. The choice of monitoring tool is a separate decision, and the comparison of an AEO and GEO agency with DIY tools covers it capability by capability.
| Point where DIY stops | What your team sees | Why its own work does not close it | What closes it | Where Aemorph does it |
|---|---|---|---|---|
| 1. The gap does not move | The same gap in the log month after month | Nobody inside the business has the capacity to ship the fix | An execution team | A monthly package, or a Sprint bought on its own |
| 2. Wrong answers persist | An engine still answers a showstopper question wrongly after the website is correct | The approved answers are not yet published in one place an engine retrieves | Approved answers published as an LLM info page, tested again after publishing | The Showstopper Sensor (Sprint 1), Aemorph's 30-day AI answer sprint |
| 3. The technical layer | Schema fragments that disagree, no entity map, thin public records | The files need one source-of-truth record, and public records are reviewed by their own communities | A connected schema graph, an entity map and edits to public records | Entity Hardening (Sprint 2), Aemorph's technical entity SEO sprint |
| 4. Regulated claims | Health or financial claims with no checking standard behind them | The team has no review step for those claims | Health claims checked against PubMed, professional associations and government health authorities; financial claims with disclaimers and no guaranteed outcomes | Aemorph's compliance practice, with client point of contact sign-off |
| 5. Scale | Many locations, services or markets to keep consistent | Each variant multiplies the facts to keep aligned | Attribute coverage built once and replicated across the variants in scope | The enterprise route, from about S$5M in revenue |
The five points where DIY stops.
Seven DIY Errors That Recur in In-House Teams
Seven errors recur in in-house work and cost visibility, and your team avoids each one without an agency. An in-house team:
- Picks topics by keyword volume, ahead of business relevance and the relationships between its facts.
- Publishes thin articles that answer one narrow query.
- Treats the work as blog publishing, with no site architecture behind it.
- Measures rankings alone, with no reading of leads or topic coverage.
- Rewrites pages repeatedly without first reading the search intent.
- Relies on AI writing tools with no subject-expert review.
- Leaves core commercial pages with no supporting content or internal links.
Two technical errors recur beside them. The first is markup for facts the page does not show; Google's structured data guidelines tell site owners not to add structured data about information that is not visible to the user. The second is blocking Google-Extended in the belief that it removes a site from AI Overviews; Google's crawler documentation states that Google-Extended does not affect a site's inclusion in Google Search.
What Hiring Help Costs Is a Separate Question
The cost of hiring help sits apart from the DIY decision, because the five points describe capacity and risk, not budget. Dated market prices from Singapore agencies and Aemorph's own starting points are set out in the AEO and GEO cost guide for Singapore in 2026.
Aemorph's Route From DIY to Hired Help
Aemorph's route from DIY to hired help, as of October 2026, matches the help to the gap in your log: it starts with two free tools, runs through Leveraged & Scalable, Aemorph's community, for businesses below about S$5M in revenue, and keeps Sprint 1 and Sprint 2 open to any business at any size. An enterprise or established business at about S$5M and above starts with a free strategic conversation.
No route is a required step before another, and a Sprint bought on its own starts with the free strategic conversation. The enterprise route runs through the Influencer Blueprint, Aemorph's SEO strategy and roadmap.
| Route | Who it is for | What it includes | Who does the work | Delivery time or guarantee |
|---|---|---|---|---|
| Free Hallucination Report | Smaller businesses; the first step below about S$5M | A crawl of the existing website and a report of where the LLMs state things that conflict with it | The tool | No call; no guarantee |
| Free Money on the Table audit | Smaller businesses; the first step below about S$5M | An estimate of the revenue the business leaves on the table | The tool | No guarantee |
| Leveraged & Scalable, Aemorph's community | Businesses below about S$5M that do the work themselves | The skills, tools and access to do the work | Your team | No done-for-you delivery and no guarantee |
| Showstopper Sensor (Sprint 1) | Any business, at any size | The Hallucination Diagnostic, the list of showstopper questions, the approved answers published as an LLM info page, a report and a call | Aemorph, with your answers and sign-off | 30 days; carries the Ground Truth Guarantee when bought on its own |
| Entity Hardening (Sprint 2) | Any business, at any size | A source-of-truth record, a linked schema graph, an entity map, the LLM info page extension, llms.txt, Wikidata edits and Google Knowledge Graph update submissions | Aemorph writes the specifications and files, makes the Wikidata edits and submits the Google Knowledge Graph updates; your developers implement the on-site files, and Aemorph verifies them | Up to 60 days; no guarantee |
| Monthly package | Enterprise or established business at about S$5M and above | A strategic conversation, an audit, the Influencer Blueprint, then Attain or Apex with Sprint 1 and Sprint 2 delivered inside the engagement | Aemorph, with your team | 45-day outline guarantee on the monthly package |
Routes from DIY to hired help.
Aemorph quotes each engagement after the strategic conversation; starting points are on the pricing page.
The Two Free Tools Come First
The two free tools are the first step below about S$5M, and both run on tools.aemorph.com. The free Hallucination Report compares the LLMs' answers with what your site currently says, and it uses no showstopper questions. The free Money on the Table audit, Aemorph's SEO Audit that counts the revenue, takes five inputs: search volume in your market, your competitors, what those competitors rank for, the expected click-through rate and your average revenue per sale.
Leveraged & Scalable Is the Recommended Route Below About S$5M
Leveraged & Scalable, Aemorph's community, is the route Aemorph recommends to a business below about S$5M that does the work itself. You supply the team time; the community supplies the skills, tools and access. Aemorph accepts a small number onto Awaken, its done-for-you package. The two revenue bands and the route open at any size are set out where Aemorph sorts Business Size into routes around one threshold.
Sprint 1 Publishes the Approved Answers, With a Guarantee
Sprint 1, the Showstopper Sensor, is open to any business at any size, and it takes over at the point where your answers are right on your website and still wrong in the engines. Your nominated point of contact signs off the answers before Aemorph publishes them as the LLM info page on your site. Sprint 1 produces no llms.txt and does not rewrite core pages.
Buy the Showstopper Sensor (Sprint 1) on its own and it carries the Ground Truth Guarantee: if ChatGPT, Claude, Gemini and Perplexity still miss your approved facts 30 days after your answers go live, you get your Showstopper Sensor fee back in full. It is Aemorph's AI accuracy guarantee. The scope is on the page for the Showstopper Sensor, the 30-day Sprint 1 engagement.
Sprint 2 Builds the Technical Layer, Including llms.txt
Sprint 2, Entity Hardening, is open to any business at any size, and it produces both the extended LLM info page and the llms.txt index. Aemorph writes the specifications and files, your developers or IT vendor implement them, and Aemorph verifies the result. Together, the two Sprints produce the Brand Codex, your brand fact sheet and the single source of truth for what AI says about you.
Treat llms.txt as optional infrastructure. It is a proposal, first published by Jeremy Howard on 3 September 2024, for a Markdown file that gives AI agents a short index of a site: the llms.txt proposal. Google states that AI Overviews and AI Mode need no new machine-readable files or AI text files (Google Search Central, updated 10 December 2025), and ChatGPT, Perplexity and Anthropic have not confirmed that they read llms.txt.
The Monthly DIY GEO and AEO Log Shows When to Bring In Help
Your monthly log shows the next step after a DIY GEO and AEO pass: close your own gaps and keep going, or bring in help at the point that matches the gap that holds back your visibility. A strategic conversation with Aemorph is free and runs 30 to 60 minutes on a video call, and Aemorph responds within 24 hours to all communication.