Currently Available: Need a skilled Software Developer for your next project?

SEO, GEO and AEO

SEO, GEO and AEO: How Search and AI Discovery Work

November 11, 2026 by Vincent Schmalbach

Start with customer search paths

Choose discovery channels by starting with the questions customers need answered. Before buying, people may ask what a product does, compare it with alternatives, check pricing, or find a nearby provider. After buying, they may need setup instructions, troubleshooting help, or account support. Each question needs different information and may lead to a different place.

Build a short map around real customer situations. A local service business might hear, “Who repairs heat pumps near me?” A software buyer might ask, “Does this tool work with our accounting platform?” An existing customer might search for steps to change a setting, cancel a subscription, or fix an error.

Include questions from first discovery through product use. A query that drives website traffic is not automatically a priority. A low-volume support question deserves attention when a poor answer causes failed setup, repeat contacts, cancellations, or lost renewals.

Use sales calls, customer interviews, support tickets, site-search terms, and chat transcripts to check whether the map reflects real needs. Ask customers how they found the business, what decision they were making, and what information they needed next. These records reveal recurring paths, though they do not capture every search or decision.

Match questions to places

For each priority question, identify where customers are likely to look for an answer. The answer may appear in standard search results, a Google AI feature, a consumer assistant that searches the web, a company chatbot, or an internal knowledge tool used by employees.

These places serve different purposes and draw on different information. A public search feature may summarize web pages and link to sources. A company chatbot may answer from connected help-center articles, product documentation, or policies. An internal knowledge tool may shape what a sales or support representative tells a customer, even when the customer never sees the tool.

Do not give every channel equal attention because it uses AI. Prioritize the places customers use for decisions that matter to the business.

If customers choose nearby providers through local search, prioritize accurate business details, service-area information, reviews, and pages that answer local service questions. If existing customers rely on the help center, improve the articles and navigation that support their tasks. If support representatives use an internal assistant, keep its connected documents current and approved.

Decide what discovery should achieve

“Be visible” is too vague to serve as a discovery goal. Define the result that matters for each customer question:

Accurate details about the business, product, price, availability, or policy
A direct visit to a product, location, support, or contact page
A brand mention in a relevant comparison
A source link that lets the customer verify the answer
A qualified recommendation when the business fits the customer’s stated need

These outcomes overlap, but they are not interchangeable. A brand mention does not ensure a site visit. A source link does not establish a recommendation. A direct visit has little value when the page fails to answer the question that prompted it.

Write the intended outcome beside each question and channel. For a product-comparison question, the goal may be an accurate explanation of features, limits, pricing structure, and fit. For a support question, the goal may be a clear answer on the relevant help page, with steps the customer can complete without contacting support.

The map shows whether to improve a product page, update a support article, correct local business information, connect better documents to a company chatbot, or monitor how a public search feature presents the business.

Do not map every possible channel

A complete inventory of queries and AI products rarely helps with the first decision. It produces a long list with no clear owner or business purpose.

Start with a small set of questions tied to a real decision: choosing a provider, comparing products, completing setup, solving a problem, or renewing a service. Then examine the places customers actually use for those questions. Expand the map when evidence shows that another path affects revenue, customer satisfaction, support demand, or the accuracy of employee answers.

Define SEO, GEO and AEO

SEO, or search engine optimization, helps search engines find, understand, and present useful pages for relevant searches. It includes making important pages accessible to search crawlers and explaining each page’s subject clearly. Google’s SEO Starter Guide describes practices that help Google understand content and help searchers decide whether to visit a result. Rankings and traffic are outcomes to measure, not results that SEO guarantees.

GEO, or generative engine optimization, is an informal label for work intended to increase the chances that a generative AI product finds or uses a business’s information. No universal definition or agreed set of GEO tactics exists. A business does not control what an AI-generated answer says, cites, or recommends.

AEO, or answer engine optimization, is also an informal label. It generally means making information easier for a search feature or answer product to use in a direct response. Practitioners use the term differently. Some use it for AI-generated answers; others include featured snippets, voice search, and other formats that answer a question before the searcher opens a result.

The labels overlap

SEO, GEO, and AEO describe overlapping work; no fixed rulebooks define them as separate disciplines. A page that search engines can access, that states useful facts clearly, and that supports claims with credible evidence serves conventional search results. It also provides clearer material for answer features and generative systems to retrieve, interpret, or cite.

Google says that the same foundational SEO practices apply to its AI Overviews and AI Mode. It also says those practices do not guarantee appearance in either feature. That guidance for site owners applies to Google Search; it does not establish rules for every AI product.

Focus on the problem that needs fixing. A business may need to make an important page accessible to search engines, answer a customer question directly, correct outdated facts, or publish information that people and systems can verify. Calling the work GEO or AEO does not identify the problem or guarantee visibility.

When a consultant or tool promises GEO or AEO, ask which search setting or answer feature it targets and which measurable outcome it expects to improve. The label does not specify the work, establish that a tactic will improve visibility, or show whether that visibility will bring qualified visitors or customers.

How an answer gets built

When a small nonprofit asks an AI search feature, “Which accounting software suits us?”, the visible answer may draw on vendor pages, nonprofit accounting guidance, pricing details, reviews, and other material. A useful general model divides the path from question to answer into four stages: discovery, retrieval, selection, and generation. Providers use different architectures, and the answer on screen does not reveal every step behind it.

Discovery finds possible information sources. Depending on the product, those sources may include web search results, a company’s private document index, files supplied by the user, or information already available to the model. A product’s name alone does not reveal which sources it consulted for a particular answer.

Retrieval fetches or identifies information for the specific request. For the nonprofit’s question, the service may retrieve pages or passages about fund accounting, donation tracking, restricted grants, reporting requirements, prices, and user limits.

Google says its AI Overviews and AI Mode may use query fan-out: expanding one question into related searches across subtopics and data sources. Its documentation on AI features, last updated December 10, 2025, describes that behavior for Google Search. It does not establish that other products use the same method.

Selection determines which retrieved information is relevant enough to include. A broad comparison of accounting products may matter less than documentation showing whether a product handles nonprofit fund accounting or restricted-grant reporting. Providers do not disclose the full selection criteria or mechanics for every product. The finished answer therefore does not show why it used one source and excluded another.

Generation uses the question and the information available to the model to compose a response. The model may compare features, summarize a vendor’s claims, and mention a limitation found elsewhere. It may also omit a relevant detail or combine claims from several sources, making the origin of each statement unclear.

A displayed link points the reader to a source, but it does not establish that the source supports every nearby sentence, that the provider endorses it, or that the linked business is the recommended choice. Check the specific passage against the claim before treating the link as evidence.

Google’s two AI search features

Google Search uses AI Overviews and AI Mode for different search experiences. The distinction matters when assessing what potential customers see: AI Overviews do not appear for every query, and the two features can return different answers and supporting links. Google defines both features in its AI features guidance.

What Google says about these features:

AI Overviews show generated summaries in selected Search results when Google determines they add value beyond classic Search.
AI Mode returns an AI-powered response with links to supporting websites for deeper exploration, reasoning, and comparisons.
A page must meet Google’s eligibility requirements to appear as a supporting link, but eligibility does not guarantee placement.

AI Overviews and AI Mode serve different searches

An AI Overview presents a generated summary with links for further reading. Google shows an Overview when its systems determine that the feature adds value beyond classic Search results. Google also says that AI Overviews often do not appear.

AI Mode is designed for questions that require further exploration, reasoning, or comparison. It returns an AI-powered response with links to supporting websites. Someone comparing accounting software for a small nonprofit, for example, might use AI Mode to weigh pricing, nonprofit features, reporting needs, and bookkeeping requirements.

Google’s descriptions apply only to these features in Google Search. They do not describe how ChatGPT, another search provider, a model API, or a private enterprise tool retrieves information or displays links.

One question can trigger several searches

Google says both AI Overviews and AI Mode may use query fan-out, in which Google expands one question into several related searches across relevant subtopics and data sources.

For a question about nonprofit accounting software, Google might search for information on pricing, integrations, donor reporting, and regulatory requirements. The visible response does not reveal every related search or the decisions that produced it.

Google also says that AI Overviews and AI Mode may use different models and techniques. As a result, the same query can produce different wording, sources, and supporting links in each feature.

Eligibility allows consideration, not placement

Google says a page must be indexed and eligible to appear with a snippet in Google Search before it qualifies as a supporting link in AI Overviews or AI Mode. Indexing means Google has added the page to its Search index. A snippet is the text description Google may show beneath a standard search result.

This requirement establishes eligibility only. Google does not promise to show every eligible page as a supporting link. The query, the page’s relevance, other available sources, and the generated response affect what appears.

Google has also said that clicks from AI Overviews are “higher quality,” meaning visitors who click through tend to spend more time on the destination site. That is Google’s observation, not an independently established result. Its documentation does not provide the underlying dataset, method, or comparison conditions, so the claim does not establish a general traffic outcome for every site.

Why eligibility does not make placement predictable

No. Meeting Google’s indexing and snippet requirements is necessary for supporting-link eligibility, but it does not create a reliable path to appearing in an AI Overview or AI Mode response. A page must still address the searcher’s specific need and compete with other indexed sources.

The practical priority remains basic Search visibility: allow Google to crawl and index the page, avoid snippet restrictions that make it ineligible, and publish useful information that directly answers the queries the page targets.

ChatGPT, APIs and company search

A model name does not reveal where a specific answer got its information. A person using ChatGPT, a developer calling a model through an API, and an employee searching company documents may receive different answers, sources, and links even when the products use related models. Evaluate each setup on its own terms.

Consumer ChatGPT search

The ChatGPT Search documentation available when this chapter was published describes ChatGPT Search as a consumer feature that brings web information into a conversation and presents links to sources. That documentation describes the ChatGPT product experience. It does not define how every product built with an OpenAI model finds or cites information.

A public page that loads in a browser has not necessarily appeared in ChatGPT Search. OpenAI might not have fetched the page or added it to a search index. It also might not have selected the page for a response or linked to it in an answer. OpenAI’s bot documentation identifies its published web crawlers, their stated purposes, and available access controls. Those details help site owners control certain automated requests. They do not show that a crawler visited a given page or that ChatGPT Search used it in a response.

APIs depend on the request configuration

An API is an interface that lets developers connect software to a provider’s services. A basic model request has its own inputs and source access. Enabling web search changes both; a request that searches uploaded files also uses different inputs and sources.

OpenAI documents web search and file search as separate tools. The developer chooses whether to use either tool and sets the request options. Those choices affect which sources the model receives and how the resulting service handles links and citations.

For example, a support assistant may search approved product manuals and retrieve relevant passages. It may then give those passages to a model to draft an answer. Another API request may enable web search instead. Neither setup necessarily matches consumer ChatGPT Search or displays the same links. Check the product documentation and the configured request rather than drawing conclusions from the model name.

Company search uses connected documents

A company can connect policies, manuals, support records, and knowledge-base articles to a search tool. For each question, the tool finds relevant passages and gives them to a model to write an answer. This approach is called retrieval: the model receives selected documents when it answers instead of relying only on knowledge learned during training.

The company controls available material by choosing connected sources and setting access permissions. An internal chatbot may answer from a private policy library that public search engines cannot access. A public web page may be absent from the company’s document collection.

A model name alone does not identify the search index, connected documents, user permissions, or retrieval steps behind an answer. Those details belong to the product and its configuration.

Test the product people actually use

Test the places where customers or employees seek information: Google Search, ChatGPT Search, a customer chatbot, or an internal knowledge tool. Use questions that reflect real needs. Record whether each product names the business, links to a relevant page, and represents the page’s claims accurately.

Repeat the tests when the product, its settings, or its connected documents change. Each result records what that product did at that time. One answer does not establish a permanent ranking rule or predict behavior in another assistant.

Also confirm that important pages load for visitors, state key facts plainly, and support claims with evidence. Crawler access is only one condition for discovery. It does not establish indexing, retrieval, citation, ranking, or business impact.

Why related models still need separate tests

Testing several products may seem unnecessary when they use related language models. Yet the product around the model determines which sources it searches, which permissions apply, and how it presents links. Shared model names do not make those conditions interchangeable.

What a source link means

A source link gives readers a visible path from an answer to a web page. A mention is a business or product name appearing in the answer, with or without a link to that business’s website. A recommendation goes further: the answer tells the reader to consider or choose the business.

These signals mean different things. An answer may mention a company without linking to it. It may link to a page without recommending the company. It may also recommend a company and include a link.

Google describes links in AI Overviews and AI Mode as starting points for readers to explore relevant content and learn more. That description applies to Google’s own search features, not every answer interface. In Google’s interface, a link indicates that Google surfaced the page in connection with the answer. It does not reveal exactly how the page shaped the wording or show that Google endorses the business. (Google’s guidance on AI features)

Check the claim against the page

A generated answer may draw on several pages, and a link beside a paragraph may support one detail, several details, or only the paragraph’s general subject. Open the linked page and find the passage relevant to the exact claim. Then check whether the passage supports the answer’s wording, including its dates, qualifications, and fit with the reader’s situation.

For example, an answer may say that an accounting product suits small nonprofits and link to the vendor’s pricing page. The page may confirm the product’s price while providing little evidence that it meets nonprofit accounting needs. The link alone does not support the suitability claim.

A source link also does not establish that the answer is accurate or complete, that the linked page takes a positive view of the business, or that the interface chose that business over alternatives. Recommendation appears in the answer’s wording. Endorsement requires evidence that the provider explicitly supports the business. Treat the link, the claim, the recommendation, and any endorsement as separate observations.

What SEO still contributes

For Google AI Overviews and AI Mode, Google says the same foundational SEO practices apply: meet Google Search’s technical requirements and policies, and publish helpful, reliable, people-first content. Google’s guidance for AI features says pages must be eligible for Google Search to be considered for these features. Eligibility and strong content do not guarantee inclusion in an AI-generated answer.

Give the relevant search engine access to important pages, and link to those pages from related parts of the site. Put the facts that matter in visible page text rather than relying on images, scripts, or downloadable files alone. If you use structured data - machine-readable code that labels details such as a product, price, or review - make sure it matches the information readers see on the page. These steps also help people who arrive through standard search results, referrals, and direct visits.

Answer the buyer’s real questions

A product or service page should explain what you offer, who it suits, and its meaningful limits. Show how it differs from relevant alternatives when those differences affect a purchase decision. Support important claims with specifications, documentation, pricing, policies, test results, or other evidence that readers can check.

This information helps customers regardless of how they find the page: through a standard search result, a link in an AI response, a recommendation, or a direct visit. When the page is relevant to a query, it also gives search engines and AI features clear material to retrieve and summarize.

People also find information about a business through sources beyond its website. Reviews, product documentation, directories, specialist coverage, and customer discussions can add facts or perspectives that the business’s own pages do not provide. When a search feature or assistant draws on web pages, those sources may shape the information it presents alongside the business’s site.

Assess external sources by relevance, specificity, and factual support. A directory listing with correct service details serves a different purpose from an unrelated brand mention. A detailed review that identifies the product, conditions of use, and result offers more evidence than many brief mentions. Mention counts alone do not establish accuracy, credibility, or usefulness.

SEO improves access, clarity, and verifiability. It helps search engines find and understand eligible pages and gives prospective customers the information they need to assess a business. The available evidence does not show a universal preference for a particular page format, answer length, markup choice, or file such as llms.txt as a route to AI citations. Fix a real customer-information or search-access problem first, then measure AI visibility as a separate outcome.

Run a focused first audit

Check ordinary search results and relevant AI search features for a small set of real customer questions. Check assistant products for those same questions, then inspect the pages on your site that should answer them and record each result with a date. A successful page request, a crawler rule, or one generated answer does not prove that a provider indexed, selected, cited, or recommended a page.

Start with real customer questions

Choose three to five questions from sales inquiries, support requests, customer conversations, or site-search terms. Use the customer’s wording; don’t turn each question into a target keyword. Include questions that name the business and questions that leave its name out.

For each question, record the product or service, relevant location, language, and search setting. A branded question might ask whether your business supports a feature. Without the brand, it could ask which providers support that feature. A support question might name a product and a specific task or problem.

Check each search product separately

Search the exact wording in ordinary search results. Then check the relevant AI features and assistant products separately. Record the date and product or mode. Note the query wording and whether the AI feature appeared. Also record visible links and business mentions.

Google states that AI Overviews and AI Mode use different models and techniques, so their answers and links can differ. AI Overviews also do not appear for every query. Record an absent feature, then repeat important checks on another date before drawing conclusions.

Do not treat a Google AI Mode result, a ChatGPT Search response, and an answer from a company chatbot as one result. Each product can use different sources and retrieval methods. One response is a sample, not a stable measure of visibility, ranking, citation, or recommendation.

Inspect the pages that should answer the questions

Identify the page on your site that should answer each question. Check whether the page is public, current, and reachable through relevant internal links. The page should state what the business offers and where the offer has limits.

For example, a feature page should identify the product or plan that includes the feature. A troubleshooting page should provide relevant steps and conditions instead of relying on a broad claim elsewhere on the site.

Treat access, indexing, and answer visibility as separate questions:

Access asks whether a visitor or crawler can request the page.
Indexing asks whether a search engine has added the page to its searchable index.
Answer visibility asks whether a specific AI feature or assistant showed, linked to, cited, or recommended the page.

A public page does not necessarily appear in search results. A page that appears in search results does not necessarily appear in an AI-generated answer.

Google’s robots.txt documentation explains how robots.txt rules guide Google crawlers. Its AI features guidance describes Google Search controls, including preview controls. Those controls apply to the Google Search behavior described in Google’s documentation. Other providers use their own crawler names and policies. Reading a site’s robots.txt file does not establish that another provider accessed a page.

For Google Search issues, use Google Search Console’s URL Inspection tool to review Google’s reported crawl and indexing information for a specific URL. Google reports traffic from AI Overviews and AI Mode under the overall Web search type. That report does not provide a separate total for AI-feature traffic.

Keep manual observations in scope

A public-site audit dated November 11, 2026 found that standard HTTP GET requests, which ask a server to return a page, to the Resources page and the Yoast sitemap audit page returned HTTP 200. Selected paragraphs from the sitemap article appeared in the response HTML.

Those checks showed what the two pages served at that moment. They did not show that a provider crawler fetched or indexed either page, or that an AI answer selected, cited, or recommended either page.

The same audit found a publicly served llms.txt response. Its presence alone does not show that an assistant reads or uses the file. Likewise, a local reading of robots.txt does not establish provider access or crawler activity. If the sitemap is the concern, use the narrower Yoast sitemap audit guide rather than turning this first pass into a technical sitemap review.

Why a small audit is useful

A focused audit identifies confirmed issues such as an outdated offer or a missing answer. It also identifies a page visitors cannot reach through internal links and a search result that points to the wrong page. It does not measure how frequently a business appears across all customer needs, locations, prompts, and dates.

Fix verified content or access problems, then repeat the same checks. Do not add a phrase or change a page format solely because one generated answer omitted the page. One omission does not establish why that answer appeared.

Measure separate outcomes

A site visit, a search result, and a mention in an AI-generated answer are separate outcomes. Record them separately when assessing visibility:

Access and indexing: whether a provider can fetch a page and has added it to its index. The provider's index stores pages eligible for retrieval.
Ordinary search presence: whether the page appears in standard search results for a specific query.
Answer visibility: whether a sampled AI answer mentions the business, links to a page, or recommends it for the question asked.
Recommendation quality: whether the recommendation fits the question and gives accurate, relevant reasons.
Visits and business results: whether identifiable visitors reach the site and whether those visits contribute to inquiries, purchases, or another defined action.

Evidence at one stage does not establish results at another. An indexed page may not appear in an AI answer. An answer may mention a business without linking to it. A displayed link does not show that anyone clicked it. A later visit or sale does not, on its own, show that an AI answer caused it.

Record only what the available evidence establishes. A crawl request, answer citation, referral visit, and purchase each support a different claim.

Treat each answer check as a dated observation, not a trend. Record the exact question, product, mode, locale, and visible sources or links. When an answer names the business, record whether it recommends the business, the reasons it gives, and whether those reasons fit the question. Repeat the same checks under comparable conditions before describing a pattern. Even repeated observations apply only to the questions and settings tested; they do not describe every answer a prospective customer will receive.

Server logs show requests that reached the site, including the user-agent string sent with each request. A request attributed to a search or AI provider shows that a client presenting that user-agent requested a resource. The log entry does not establish that the provider indexed the page, used it in an answer, ranked it prominently, or influenced a business result. User-agent strings identify themselves rather than proving the requester’s identity.

Likewise, an answer link shows that the answer displayed a page as a source. It does not show whether a reader opened the page, read it, agreed with the answer, or later became a customer.

Use the same question sample for each comparison. If one review contains mostly branded questions and the next contains mostly generic comparison questions, the mention or link rate may change because the questions changed. Record added and replacement questions separately from the original sample. Compare like with like before attributing a difference to improved visibility.

A combined metric produces one easy number, but it hides where the evidence ends. A higher answer-mention rate, for example, does not establish more clicks, qualified visits, or revenue. Separate measures require more careful records, yet they preserve the difference between observed behavior and inferred impact.

For Google, Search Console includes traffic from Google’s AI features within overall Web search reporting. That reporting does not isolate every step between page eligibility, an AI-generated answer, a visit, and a customer decision. Measure what each record shows and leave unsupported causes unknown.

Avoid GEO promises

Generative engine optimization, or GEO, refers to work intended to increase a business’s visibility in AI-generated answers. It deserves attention when customers use AI search or an assistant to compare products, find providers, or assess claims. Businesses should make their information accurate, accessible, and easy to verify for those questions. That work also improves conventional search and helps customers understand the business when no AI product links to it.

Scrutinize any separate GEO proposal that promises guaranteed citations or recommendations, prescribes a fixed answer format, or claims that one technical file controls every AI product. Ask which product and customer questions the proposal addresses, what documented provider behavior supports its recommendations, and how the supplier will measure the result. A supplier can audit content, repair pages that search engines cannot access, and publish well-supported product facts. It cannot control another company’s generated answer.

Google’s guidance on AI features in Search, updated December 10, 2025, says a page must be indexed and eligible to appear as a Search snippet before it can qualify as a supporting link in AI Overviews or AI Mode. Eligibility does not guarantee inclusion. Google also states that AI Overviews do not appear for every query. Those rules apply to Google Search. Other AI products document different behavior, and many disclose less about how they retrieve pages or choose sources. Without that disclosure, an observer cannot determine why a page appeared or failed to appear.

Do not treat factual accuracy, cited links, and brand recommendations as one outcome. A third-party page might mention a business while stating a product fact incorrectly. An AI answer might link to a page that supports one narrow detail but not its broader conclusion. A business with useful, well-supported pages might still be absent from an answer because the provider selected different sources, interpreted the question differently, or did not generate an answer for that query.

Fund work against a diagnosed problem. Customers may be unable to reach a key page, product details may be outdated or incomplete, or important claims may lack credible evidence. For example, a manufacturer that has specifications only in a downloadable image should publish those specifications in readable page text. A local provider whose service area or pricing rules are unclear should state them plainly and keep them current. These changes create a clear deliverable whether or not a particular AI product cites the business.

Once the work is complete, review relevant search results and named AI features against a defined set of customer questions. Report what changed, what did not change, and what the provider does not disclose. Observed appearances and citations provide useful evidence about those specific tests. They do not prove a durable ranking position or guarantee future inclusion.

Key takeaways

Make customer-facing facts accurate, accessible, and easy to verify across discovery channels.
Fund defined work that fixes a known access, content, or evidence gap.
Reject guarantees about citations, recommendations, or universal control over AI-generated answers.
Judge results by observed outcomes while stating the limits of what those observations show.

A dedicated GEO program has a defined use when it gives owners a disciplined way to test how customers encounter the business through named AI products. The label creates problems when it substitutes a universal recipe for product-specific evidence. Set a narrow scope, such as improving answers to five high-value customer questions in Google AI Mode or another named service. Measure visible results without claiming control over the provider’s ranking or answer generation.

Key Takeaways

Start with customer questions: Find where customers look for answers, then prioritize the search features and assistant products that matter to your business.
Treat SEO, GEO, and AEO as overlapping labels: Improve access to accurate, useful information rather than relying on a separate formula for each term.
Separate discovery from results: A page can be accessible, indexed, retrieved, linked, or mentioned, and each status describes a different event. A source link alone does not prove that the answer supports every claim or recommends the business.
Audit specific gaps: Check relevant search results and AI products separately, then investigate outdated content, unclear claims, and access or indexing issues. Record the date and conditions because one answer is only a snapshot.
Reject visibility guarantees: No universal GEO tactic guarantees a citation or recommendation. Fund work tied to a diagnosed problem and evaluate visibility, visits, and business outcomes as separate measures.

FAQ

Does GEO replace SEO?

No. GEO is an informal label for increasing the chance that generative systems find or use a business’s information. AEO commonly refers to making information useful in direct-answer formats. The labels overlap, and Google says its AI Overviews and AI Mode use the same foundational SEO practices as Google Search.

Does resisting GEO promises mean ignoring AI search?

No. Monitor the AI products customers use. Correct information gaps that prevent accurate discovery, and test changes where results are visible. Resist claims that turn provider behavior the business cannot control into a guaranteed deliverable.

Does appearing in Google Search guarantee inclusion in an AI Overview?

No. Google says a page must be indexed and eligible to appear with a snippet in Search to qualify as a supporting link. Even then, it may not appear. Google also says AI Overviews do not trigger for every query.

Does a link in an AI answer mean the business is recommended?

No. A source link gives the reader a route to a page; a recommendation advises the reader to consider or choose a business. The link alone does not show that the page supports every nearby claim or that the provider endorses the business.

Does a citation in an AI answer prove that the cited page influenced the answer?

It proves that the answer displayed the page as a source. It does not show how much the page influenced the wording, whether the reader opened it, or whether the reader acted on the answer.

Does an `llms.txt` file guarantee that AI assistants will use a site?

No. A public response from that file proves only that the server returned it for the request. It does not establish that a provider reads the file, retrieves the site, or uses its content in an answer.

How do I start checking AI discovery for my business?

Choose a small set of real customer questions, then check ordinary search and each relevant AI product separately. Record the date and exact question for each check. Also note which product or mode you used, whether an AI feature appeared, and any visible mentions or links. Treat each result as an observation rather than a stable ranking.

What if my page returns HTTP 200 but does not appear in an AI answer?

HTTP 200 confirms that the server returned a successful response to the request. It does not establish whether a provider crawler fetched the page, a search engine indexed it, or an AI system selected it. Check access, indexing, and the page’s content separately.

Does a crawler visit prove indexing?

No. A crawler request proves that the server received a request carrying that user-agent string. Indexing requires separate evidence from the provider or observed search presence.

Can I tell exactly why an assistant chose one source?

Usually not from the answer alone. The visible response does not reveal every search, source, selection step, or model decision, and providers publish different amounts of information about their products. A model name by itself also does not identify the search index or connected documents used.

What I'm building

Delegate tasks. Get software.

Give Vroni a GitHub issue, bug report, spec, or rough idea. It reads the repo, plans the change, writes code, runs checks, and works toward a review-ready pull request.

Take a look at vroni.com