// geo guides
Answers to common questions about visibility in AI search. Each guide starts with a short answer, explains the detail and lists its sources.
In the US and Canada, ChatGPT's map card usually shows the Yelp rating. Where Yelp and Google ratings differed, the card matched Yelp in 55 of 68 cases in Houston and Miami on 27 September 2026, and in 29 of 31 in Chicago and Toronto the next day. In London, Manchester and Sydney it mostly matched Google.
To prepare your company for AI search, make key pages accessible to crawlers, describe your services clearly, support claims with verifiable evidence and repeat your buyers’ questions regularly. This work on visibility in AI answers is called GEO. You can delegate technical work, writing and measurement; your company remains responsible for facts, expert input and permission to disclose client information.
An e-commerce AI visibility audit records answers to customer questions, identifies cited sources and checks product details for accuracy. The report should propose specific fixes, each with an owner and a verification step. Count brand mentions, links and shopping offers separately. Below is an example report.
A new business can appear in ChatGPT recommendations, but being found by name does not demonstrate that customers will discover it. Start with a specific offer you can substantiate, publish the facts a buyer needs and test relevant questions without your brand name. Use repeated observations and real enquiries to decide whether to expand.
Generative Engine Optimization (GEO) is the practice of structuring a brand and its content so generative AI systems — ChatGPT, Perplexity, Gemini, Claude and Google AI Overviews — discover it, trust it and cite it inside generated answers. Where SEO competes for a ranked link, GEO competes to be the source the model quotes.
SEO helps a website rank in search results and earn clicks. GEO helps its content become a cited source in an AI-generated answer. Both rely on crawlable pages and structured data. They differ in how content is organized, what counts as success and how results are measured.
A model’s behavior depends on the system around it: search, tools, instructions, ranking and safety controls. The same model can therefore cite different sources in different products. GEO addresses how each product finds and reads pages, and which external sources support the information on them.
ChatGPT doesn’t recommend a company for one of seven reasons: its crawler can’t fetch the page, no page answers the buyer’s exact prompt, no passage can be quoted, the brand is described inconsistently across sources, the company is absent from “best X” lists, the page looks undated, or the model checked the official site and couldn’t confirm the claim.
How long GEO takes depends on which of three layers you fixed. Access and rendering fixes reach AI answers fastest, in days to weeks. New pages and consistent brand naming take weeks to a quarter to show in repeated runs. Independent brand coverage moves over quarters. On our own site, launch to first reproducible ChatGPT citations took about two months.
Mostly, yes. AEO, GEO, LLMO and AI SEO share a goal: becoming a source an assistant uses in its answers. AEO (2017–2018) grew out of featured snippets and voice search; GEO (2023) is the research term for citations in generated answers; LLMO focuses on what the model already knows. The work largely overlaps, but the measures differ.
Watch for promises you cannot verify: guaranteed mentions, “submission” to ChatGPT, an unexplained visibility score, a hidden prompt list or a case study that conceals the agency’s ownership of the site. For each warning sign, this guide gives a question to ask and an example of a clear answer.
GEO works at the level it can be measured. A page written for a specific buyer prompt, on a site AI crawlers can read, gets cited on that prompt within weeks to a quarter, depending on the access work needed. Brand-level lift on generic prompts in crowded categories is not proven, and anyone claiming it skipped the measurement.
Sometimes. A paid spot in a list article helps only if AI assistants already use that page as a source and the listing links to a page of yours that answers the buyer’s question. If the seller cannot show either, you are buying link building under a new name, with the old link building’s uncertainty attached.
GEO costs $2,500–6,000 for our 90-day pilot, $750–1,500 for a deep audit, $1,500–3,000 a month on retainer and $5,000–15,000 for a GEO-ready website; the first AI visibility audit is free (USD, September 2026). Each tier is a public offer with a fixed scope. In our September 2026 check, 47% of AI-cited GEO agencies published prices; median entry: $3,250 a month.
To be cited by ChatGPT you need three things: crawler access for OAI-SearchBot, the only OpenAI agent that decides whether a site appears in search answers; server-rendered pages where one question's answer comes first and contains a checkable fact; and corroboration, independent sites describing you the same way. Access takes a day, pages weeks, corroboration quarters. Placement is never guaranteed.
To get cited by Perplexity, a page has to clear three bars: PerplexityBot can reach and read it; it states an answer the model can lift directly, backed by specific, verifiable facts; and it is corroborated beyond your own domain — by classic search rankings and independent mentions. Everything else in “Perplexity SEO” is a variation on these three.
Claude can mention your brand two ways, and only one is measurable: live web retrieval via ClaudeBot. Training-data mentions are opaque. To win citations, allow ClaudeBot in robots.txt, ship extractable answer-first pages, and earn corroboration on sources Claude already trusts — then verify with a fixed question set.
AI Overviews are generated from Google's regular search index, so appearing in one starts with ranking for the query. Beyond that, Google documents no separate opt-in: the levers are conventional indexing, allowing the Google-Extended token, structured data, and content written as compact self-contained answers that are straightforward to summarize and attribute.
Measure AI visibility with four inputs: citation rate across a fixed prompt set sampled repeatedly, share of voice against competitors on those same prompts, AI-crawler hits in server logs as the earliest signal, and referral traffic from assistant domains as the commercial outcome. Single checks are unreliable because answers vary between runs.
A half-day audit of get-geo.ai found conflicting apex and www settings, an outdated suggestion from an AI agent and incomplete text exports. We aligned the canonical URLs, sitemap and llms.txt on one host. Before publishing more content, check that crawlers can reach the right version of what is already there.
About half of US adults use chatbots in 2026. Fewer people click search results when an AI summary appears, and the sources AI systems cite change from month to month. The figures below, with their sources, help you assess the importance of GEO for your business and decide how to measure it.
We publish our GEO measurement method: a fixed set of questions, repeated tests in clean sessions, defined citation and share-of-voice metrics, and records you keep. After 90 days, if measured visibility has not improved, we explain the findings and let you choose how to proceed.
Ask for evidence rather than process: an anonymised before/after dataset from a real client, prompt counts per language, how mentions are told apart from recommendations, how share of voice is computed, what happens if visibility rises and revenue does not, and who actually does the work. Below are the ten questions, what a good answer contains, and our own answers.
Israeli buyers use Hebrew and English, often during the same purchase. Assistants draw on different sources in each language, so visibility in English may not carry over to Hebrew. Keep business information consistent, publish natural Hebrew pages that answer questions directly and measure results in both languages.
Buyers in India ask questions in English, Hindi and Hinglish. Assistants may treat the Devanagari and Latin forms of a brand name as different businesses. GEO here requires clear content in both languages, consistent names and contact details across scripts, and separate tests for the wording buyers use.
Swiss GEO requires consistent business information in German, French and Italian. The same question can produce different answers in each language. Publish natural local versions with matching facts and reciprocal hreflang tags, then measure a separate question set for each language instead of relying on an English-only test.
A US GEO plan should test English and Spanish separately. Build useful pages, keep business information consistent and examine the independent sources assistants cite for your category. Choose languages using customer evidence, then track links and mentions over repeated checks. Fewer Spanish sources alone do not establish demand or promise faster results.
GEO in Dubai serves buyers who ask questions in English and Arabic, often during the same purchase. It requires useful content in both languages, consistent names and contact details, and crawler access. Thin Arabic pages or conflicting business details make it harder for assistants to identify and recommend the company.
Multilingual GEO, also called AEO, aims to earn citations in each language your customers use. Assistants draw on different sources for different languages, so English results do not establish Hebrew or Hindi visibility. Publish natural pages with clear answers, keep brand information consistent across scripts and measure a fixed question set for each language.
Russian-language GEO — also called AEO, Answer Engine Optimization — targets two separate arenas: Russian-speaking buyers abroad, in Dubai, Israel, the EU and Central Asia, who ask ChatGPT, Perplexity and Gemini; and users inside Russia, where OpenAI restricts access and Yandex-ecosystem assistants dominate. Most brands win first in diaspora prompts, where commercial Russian coverage is thin.
GEO in Hindi — also called AEO, Answer Engine Optimization — targets the widest gap in AI search: roughly 600 million speakers served by less than 0.1% of web content. Native Devanagari answer-first pages, entities declared in both scripts, and Hinglish prompt coverage let one well-structured source own Hindi answer slots in India and across the diaspora.
German GEO, also called AEO (Answer Engine Optimization), depends heavily on independent sources that support a brand’s claims. German accounts for about 5.9% of web content, giving assistants many sources to compare. Priorities include natural German pages that answer questions directly, consistent business information and accurate profiles on relevant review platforms.
French GEO — also called AEO, Answer Engine Optimization — is about being cited when assistants answer in French. Nearly half of France already uses generative AI, and the market adds a domestic engine, Mistral's Le Chat, worth measuring alongside ChatGPT, Perplexity and Gemini. Native French pages with aligned facts across every francophone market win the citation.
Italian GEO, also called AEO (Answer Engine Optimization), helps brands appear in Italian AI answers. Italian accounts for 2.8% of web content, with fewer competing sources than German or French. Focus on natural Italian pages, clear answers and consistent brand information. Measure a fixed set of questions, including relevant city variants, against a starting baseline.
Spanish is the web’s second-largest content language, but each market has different sources. Questions about Spain and Mexico can produce different recommendations. Spanish GEO, also called AEO, therefore needs test questions for each country, natural local vocabulary, consistent information on relevant websites and separate measurement by market.
Chinese GEO — also called AEO — targets Chinese-language answers on ChatGPT, Perplexity and Gemini, none of which is officially available in mainland China. The winnable market is the open Chinese web: Taiwan, Hong Kong, Singapore, Malaysia and the global diaspora, where content is thin enough that one structured source in both scripts can dominate the citation slot.
SaaS buyers now ask assistants for category, comparison and alternatives recommendations, and engines answer from review platforms, documentation and pricing pages they can extract. GEO — also called AEO, Answer Engine Optimization — makes those assets citable: honest comparison pages, public docs, transparent pricing, one consistent entity, measured against a fixed prompt set.
Home buyers ask AI assistants where to buy, whether to rent or buy, and which developments are open to international buyers. Answers draw on market reports, area guides and property portals. GEO, also called AEO (Answer Engine Optimization), helps agencies and developers become sources through useful local pages, verifiable figures, consistent business information and measurement by city and language.
GEO for clinics — also called AEO or LLM SEO — is not about outranking WebMD on symptoms; assistants route medical questions to institutional sources. Clinics win the commercial layer: named packages with prices, verifiable accreditations, doctor credentials and availability, published as structured citable facts and measured against a fixed prompt set.
GEO for e-commerce helps stores make their products easier to find and describe in AI answers. The work combines accessible product facts, useful buying advice, verifiable claims and appropriate catalogue integrations. Measure recommendations, citations and shopping offers separately, then connect them to visits and sales where attribution is available. Placement is never guaranteed.
Setting up robots.txt for AI crawlers means making three separate decisions, not one: whether to admit training crawlers, search-index crawlers, and user-initiated fetchers. Every vendor runs different user-agents for each job. Most sites should allow search and user fetches, decide deliberately about training — and know that some fetchers ignore robots.txt entirely.
llms.txt is a proposed markdown map of your key pages for AI systems, placed at /llms.txt. No major vendor has committed to reading it, and most published files get zero requests in a given month. Publish one if it costs nothing to generate; expect nothing from it yet.
The Gemini app, AI Mode and AI Overviews are separate services. For Google Search’s AI features, pages must be indexed, eligible for snippets and included through Search Console. Publish useful answers with first-hand evidence, then track each service separately. Google does not require llms.txt, special content chunking or an “AI writing” format.
A law firm gets recommended by AI assistants through verifiable commercial facts, not articles about the law: one page per practice area and jurisdiction, a stated fee model, attorneys with bar admissions, consistent directory profiles, measured against a fixed prompt set. Hiring prompts with a city and practice area are open ground; legal-information prompts go to courts and bar associations.
If citation rate and share of voice have not moved against your day-one baseline after 90 days, we tell you first, with our analysis of why. You then choose: a free diagnostic month, a change of scope, or a stop. The baseline, the prompt set, the logs and the reports stay yours in every case.
A bank or fintech gets recommended by AI assistants through verifiable commercial facts: one page per product, segment and country with dated fees, eligibility and licence; a consistent entity on Wikipedia and comparison sites; wording that passes compliance; all measured against a fixed prompt set. 55% of Americans now use AI for personal finance decisions (TD Bank, February 2026, N=2,504).
Base the decision on available time and skills. If one person has four to six hours a week, can edit the site and serves buyers in one language, start in-house: five steps, the first four in about one working day. An agency earns its fee on pages in several languages, independent sources and repeated measurement.
Check it by hand: write 10 prompts your buyers would ask (not “tell me about us”), run each 2–3 times in clean sessions on ChatGPT, Perplexity and Gemini, and log who was named and which URLs were cited. About two hours. One run is a snapshot, not a diagnosis — absence has three different causes, each with a different fix.
A GEO contract should fix nine things in writing: the scope by layer, a prompt set of 30–50 questions per language that you hold a copy of, two defined metrics, a reporting cadence, what happens at day 90 if nothing moved, the notice period, and a data clause stating that the baseline, logs, reports and pages are yours.
No GEO agency can guarantee a citation in ChatGPT, Perplexity or Gemini: OpenAI states that placement is not guaranteed, and answers vary between runs. What can be guaranteed is the process. Ours is a formula: 30–50 prompts per language, clean sessions, remeasurement every 2–4 weeks, and at day 90 a review with three options and data that stays yours.
Check access, rendering, content, company information and measurement in order. Can search crawlers reach the final URL, and is the answer available in the HTML? Then compare repeated tests with your baseline. OpenAI’s roughly 24-hour robots.txt update window is not a recrawl deadline; missing log entries alone do not prove an access fault.
Decide by workload and available skills. With one language and up to about 30 buyer questions, start in-house and add a tracker if useful. Hire an agency for languages or independent-source outreach your team cannot cover. Someone still needs to write, implement fixes and interpret the results; a subscription alone does not complete that work.
Mostly the pages of the companies that sell what is being asked about. In our September 2026 count of 569 source links from ChatGPT, Perplexity, Google AI Mode and the Gemini app, 48% pointed to vendors' and agencies' own sites, 16% to directories and comparison sites, 10% to forums, 7% to platform documentation and 4% to media.
ChatGPT mostly recommends the law firms shown on its map of local business listings, not the firms whose websites it cites. In our study of 72 answers on 22 September 2026, it showed a map in 11 of 12 lawyer searches and cited directories and bar associations. Perplexity cited firm websites far more often.