A customer asks a question and receives an answer assembled from several sources before opening a single site. The route now looks like this: question → AI answer → sources → brand check → site → conversion. The answer already carries the company name, its services, prices, terms, and an opinion about it, so part of the first impression and the vetting happens before the click, and the link to the site becomes one of the next steps.
A company controls four things on that route: the availability of its pages, the accuracy of the facts about itself, the completeness of those facts, and their consistency across public sources. Measuring conversion after the click and the composition of the digital asset are covered in separate articles; this one stays within what the company can control.
What follows shows which sources these systems can draw company information from, which technical conditions can be checked, which facts have to be consistent, and how to observe brand presence in answers without promising citation. Different AI systems follow different rules, so every statement here is tied to a specific system or to an observation.
An AI answer becomes a separate point of contact
Google Search has two AI formats: AI Overviews and AI Mode. The second extends the first with a more interactive answer that links to web resources. Separately there are chat assistants such as ChatGPT and Perplexity, which answer questions and cite sources under their own rules. These environments run on different mechanisms, and merging them into one model produces a wrong picture of each.
They share one thing: the answer to a complex or follow-up question is assembled from several documents, and the user can continue the dialogue instead of clicking through. The moment of first contact with a brand moves from the site into the answer. As industry observers describe it, the main product of such an answer is the synthesized text, and a link can be prominent, one of many, or hidden until expanded.
For an owner this means a new point of information control: what a system says about the company depends on what it finds and on how consistent those findings are.
Sources shape the public picture of a business
The company’s own site
The site is the only source a company controls in full: products, services, expertise, terms, contacts, original material. Every fact on the site has a person responsible for keeping it current and a date of the last check; the question of who owns the site and search visibility as an asset is covered in a separate article.
The search index
For Google, a page can appear in AI Overviews or AI Mode when it is indexed and eligible to be shown with a snippet. The index holds pages, snippets, images, and video; whatever is outside the index stays outside Google’s answers.
External sources
Company profiles, directories, publications, reviews, and partner platforms exist outside the company’s control. They are treated as an environment to observe: which facts sit there, how closely those facts match the site, and who is responsible for updating them. This article makes no claim that external sources directly affect inclusion in any particular system’s answer: for none of these systems is that documented.
The source map has four columns: source → which facts → who keeps them current → how often they are checked. There is no universal list of sources that covers every AI service; for systems without documentation, the map is filled in from observation of a control sample.
Technical availability decides whether a page takes part
Google’s position is documented. There are no extra requirements and no special optimizations for AI Overviews and AI Mode; a page has to be indexed and eligible for a snippet; structured data has to match the visible text; for online stores and local businesses, Merchant Center and Business Profile have to be current; even when every condition is met, appearance is still not guaranteed. Control over how snippets are displayed works through the usual tools: nosnippet, data-nosnippet, max-snippet, noindex.
Google has also documented what to leave alone: Google Search makes no use of files such as llms.txt, of special AI files, of separate markup, or of Markdown, and files of that kind have no effect on visibility in Google. For other systems the function of similar files is equally undocumented, so spending on them has no confirmed purpose.
The technical conditions that technical SEO and development check for key pages: crawl availability, indexability, snippet eligibility, key information in text form, internal links to the page, page experience, and structured data that matches the visible text. Page experience metrics (loading, responsiveness, and layout stability) are covered in our article on Core Web Vitals.
Since 2026 Search Console has carried a setting that the owner decides on. The Search generative AI control toggle has three states: include the site’s links and content in Google’s generative AI features (the default state), exclude them, or inherit the setting from the parent property. Exclusion removes the site’s links and content from AI Overviews, AI Mode, and the generative features of Discover, and it also blocks the use of the content in forming an answer; the toggle leaves the rest of Search untouched and does not affect ranking; the change takes effect within a few days. The business consequence: exclusion closes the site’s access to the demand that flows through these features, and inclusion keeps the company on the route, though appearance is not guaranteed. The decision follows from the business model and is made deliberately, with the date and the reason recorded.
For other systems, only the bot access mechanisms are documented. OpenAI uses OAI-SearchBot to show sites in ChatGPT’s search features, GPTBot to collect data for model training, and ChatGPT-User for user-initiated actions; the first two are managed through robots.txt, while robots.txt rules may not apply to the third. Perplexity uses PerplexityBot, which respects robots.txt and is not used for model training, and Perplexity-User, which usually ignores robots.txt for user-initiated requests. That is the full extent of what is documented; everything else about the behavior of these systems is recorded through observation.
An owner leaves this section with a list of technical conditions to check and one management decision on the site’s participation in Google’s AI features.
The structure of the data ties facts to the company
Company and brand
The public and legal names, the address, the geography of operations, contacts, and official profiles. A discrepancy between the name on the site, in the register, and in a profile produces two different companies in a system’s answer, or one company with the wrong contacts.
Products and services
Products and services, categories, specifications, prices, availability, and purchase terms. Every fact has a source page on the site, and it travels to other sources from there.
Authors and experts
Credentials, date, editor, and the method behind the material. Google recommends author bylines where the reader expects them, and a link from the byline to information about the author; Google sets no target length for that text.
Structured data plays a supporting role: it has to match the visible text, and Google states that generative search does not require it and that there is no special schema for it, while general SEO practice still finds it useful. This article promises no citation through markup, because the documentation contains no such promise.
All these facts come together in a register of key entities with four columns: fact → source page → owner → date of update.
Evidence-based content creates a source
A system that assembles an answer from several documents needs documents worth drawing on. Original research, methods, industry analyses, primary data, precise answers to hard questions, documentation, and expert commentary give a company the role of a source. Generic content available on ten other sites gives it a different role.
Google describes this as “non-commodity” content carrying the unique experience of an expert, and at the same time lists as unnecessary the splitting of text into small chunks, the rewriting of material “for AI,” artificial mentions on third-party resources, and excessive attention to structured data.
For every evidence-based publication the author, the method of obtaining the data, the limitations, and the date of update are recorded. The material is addressed to the person who makes the decision; material written for a machine loses both the human and the machine. A content plan on this logic is built around the company’s own evidence and around the decisions its audience makes.
Brand consistency reduces contradictions
A contradiction between sources produces an inaccurate answer about the brand, and the customer sees that answer before the click. What gets checked: the name, the description of the business, products, prices, terms, geography, contacts, experts, dates, legal details, and official profiles. Every fact gets an owner and an update cycle.
Reviews and mentions enter the check as a public environment: the company knows what is written about it and corrects factual errors where that is possible. The effect of reviews on an answer stays undocumented; this article treats them as an observable risk. The company comes out of the consistency check with owners assigned to the facts and an update cycle set.
An online store passes on current product data
For an online store the sources of product truth are the product pages, the categories, Product structured data, the product feed, and Merchant Center. Price, availability, delivery, returns, and product variants have to match across all of them at one point in time. Google recommends setting up automatic updates of Merchant Center data from the site’s content, so that price and availability stay in agreement between the site and the feed.
The business risk of a discrepancy is concrete: the customer sees the feed price in the answer, clicks through to the site, and finds a different one. What follows is an abandoned purchase, a return, or a complaint. The owner receives the list of product truth sources and the schedule for synchronizing them; markup fields, implementation code, and integrations with the accounting system that supplies prices and inventory stay with development.
AI presence calls for its own measurement method
Google Search Console data
Since June 2026 Search Console has carried a separate Generative AI performance report for Search. It shows impressions of the site’s pages in AI Overviews and AI Mode by page, country, device, and date; it carries no clicks, queries, or CTR; Search Labs data is excluded; Google plans to extend the list of features; the report forms part of the “Web” search type data in the general performance report; since August 31, 2026, it has been available to every site, and the usual 1,000-row limit applies to it. Google warns that the freshest data can be preliminary.
Clicks on links from AI Overviews and AI Mode count in the general performance report as ordinary clicks, with no separate breakdown. The consequence for an owner: through Google it is possible to observe impressions in AI features and the overall click trend, while clicks specifically from AI answers stay unknown at page level. The share of AI-driven visits through Google is unmeasurable today, and this article says so plainly.
Visits from other AI services
For systems outside Google, the source of a visit is identified by the referral domain in web analytics. For ChatGPT, links observed in analytics often carry the tag utm_source=chatgpt.com. For every recognized source the landing pages, the actions, and the quality of the visits are recorded. This is the one way to see visits from chat assistants to the site, and it works for the sources that pass a referrer.
A control sample of answers
Observing what systems say about a brand calls for a reproducible protocol. The company approves a fixed set of questions about itself and repeats it monthly in each system, recording the conditions: date, system, region, language, whether the session was signed in. For every question the record holds the answer about the brand, the sources used, the factual accuracy, and — for systems with a recognizable referral — the visit and the conversion, measured as described in the article on the route from click to sale.
| Query | System | Date and conditions | Answer about the brand | Sources used | Factual accuracy | Visit | Conversion |
|---|---|---|---|---|---|---|---|
| [illustrative example] “Who installs boilers with a warranty in Chicago” | Google AI Mode | September 5, 2026, United States, English, signed out | Company, service, and city named; price absent | Service page, aggregator directory | Name and city correct; phone number out of date | Page-level data unavailable: the report holds impressions only | Page-level data unavailable |
| [illustrative example] the same query | ChatGPT | September 5, 2026, same conditions | Company named; warranty terms restated with an error | Service page, blog article | Warranty period stated wrongly | 3 sessions in a month, referral chatgpt.com | 1 inquiry |
Answers vary from query to query and from day to day, so the sample describes the state as of the date of the check. It works as a record rather than an absolute ranking, and a single query is a one-time snapshot rather than monitoring. The terms GEO, AEO, and LLMO used by the market denote the same work on availability, data, and content; as a separate technology with a confirmed mechanism they remain undocumented.
A log of observations, with honest gaps in the data and measurable consequences where those are available, is the method a company can reproduce every month.
A 90-day plan
Days 1–30: inventory
The register of company entities, the list of key pages and sources, a check of indexing and snippet eligibility, a list of contradictions between sources, a fixed set of control questions, and a decision on the Search generative AI control toggle with the reason recorded.
Days 31–60: implementation
Corrections to the structure and the data on source pages, preparation of evidence-based materials with authors and dates, synchronization of product feeds, and the setup of observation: the Generative AI report in Search Console and referral sources in web analytics.
Days 61–90: verification
A repeat control sample, analysis of sources, impressions, visits, conversions, and factual errors, and the next priority added to the backlog.
The plan sets the order of the work and the checkpoints. It promises no inclusion in answers within 90 days and sets no target values for impressions.
SEO and development build the foundation of AI presence
The cycle starts with the facts about the company: business data → site structure → technical implementation → content → presence check → conversion. The register of entities defines what has to be on the site; the structure and the technical implementation make it available; evidence-based content gives the systems a source; the control sample and the reports show the result; conversion after the click is measured by the rules given in the section on the measurement method.
Inward Labs prepares the structure and the content of a site so that systems recognize the company, its products, services, and expertise correctly, and does that within the overall SEO architecture. AI presence is managed by the same technical and content system as ordinary search, so it works within the existing program rather than as a separate “AI project.”
What the owner should do
- Assign owners to the facts about the company according to the register of entities. Owner: head of marketing. Data: the register from the section on the structure of the data. Timeframe: two weeks.
- Check indexing and snippet eligibility for the key pages. Owner: the SEO specialist together with development. Data: Search Console. Timeframe: two weeks. Inward Labs expertise in technical SEO covers checking source pages.
- Make the decision on the Search generative AI control toggle and record the date and the reason. Owner: head of the company. Timeframe: one month.
- Approve the fixed set of control questions and the monthly cadence of the sample. Owner: marketing. Timeframe: one month.
- Set up observation of referral sources from AI services in web analytics. Owner: the analyst. Timeframe: one month.
CHECK YOUR BUSINESS PRESENCE IN AI SEARCH
Inward Labs will check the availability of the site, the consistency of the data, the sources about the brand, and the readiness of the key pages for search and AI answers.
