This glossary explains 60 terms of AI search in plain English: what the work is called, how ChatGPT, Gemini and Google's AI build an answer, how to measure it, which crawlers to let in and which signals AI trusts.
Where we have our own numbers, we say so. Most come from our study of 500 real buyer questions in English and Arabic, run on 21 October 2026. Outside facts are linked in the sources at the end.
The names for the work
AI SEO
AI SEO is the work of getting your business named when people ask an AI tool who to choose: ChatGPT, Gemini, Google's AI Overviews and AI Mode, Claude, Perplexity or Copilot. It is also what UAE buyers call it: in our Google Trends pull for October 2024 to October 2026, "AI SEO" averaged 13, while "generative engine optimization" and "answer engine optimization" averaged 0. Start with our AI SEO guide.
GEO (generative engine optimisation)
GEO is another name for AI SEO, from a 2024 paper by researchers at Princeton and partners. It found that adding cited sources, quotations and statistics raised a page's visibility in AI answers by 30% to 40% on its main measure, while keyword stuffing did little. In our October 2026 scan of the UAE market, about 39 of 53 search specialists (74%) sold GEO or AEO. More in GEO vs AEO vs AI SEO.
AEO (answer engine optimisation)
AEO is a third name for the same job, from the idea that AI tools are "answer engines" that give one answer instead of ten links. Ask an agency what work it does, not which label it uses.
LLMO and AIO
Two more labels: large language model optimisation and AI optimisation. From what we have seen on agency sites, these are vendor and agency names for the same work as GEO, with no separate method behind them.
Large language model (LLM)
An LLM is the kind of AI model behind ChatGPT, Gemini and Claude. It learns patterns from a huge amount of text, then writes answers word by word. On its own it only knows what was in that text, which is why assistants also search the web when a question depends on current prices, places or businesses.
Where AI answers appear
AI Overviews
AI Overviews are Google's AI summaries at the top of some results, with links to the pages used. Google said in July 2025 that they had over 2 billion monthly users in more than 200 countries and territories and 40 languages, and they reached the Middle East and North Africa, with Arabic, in May 2025. They appear only when Google judges them useful. See our AI Overviews guide.
AI Mode
AI Mode is Google's conversational search: a separate tab where people ask longer questions and follow-ups, and Google answers with links. Google said it passed 1 billion monthly users in May 2026, one year after launch, and it has supported Arabic since 8 October 2025. Our AI Mode guide covers how it picks pages.
Gemini app
The Gemini app is Google's chat assistant at gemini.google.com and on phones, with more than 1 billion monthly users (Google, August 2026). It answers either from the model's memory or grounded in Google Search, and Google says not all of its responses include sources. Details in our Gemini guide.
ChatGPT search
ChatGPT search is what happens when ChatGPT looks things up before it answers. OpenAI says it rewrites the question into one or more searches, sends them to third-party search providers and answers with links. OpenAI reported more than 1 billion weekly ChatGPT users on 31 August 2026. Sites that block its search crawler are not shown. See our ChatGPT guide.
How AI builds an answer
Training data
Training data is the text a model learned from before it was released. Crawlers such as GPTBot and ClaudeBot collect public pages that may be used for this. What a model learned only changes when a newer model is trained, so it lags behind your current prices, address and services.
Memory answer vs live-search answer
A memory answer comes from training, with no links. A live-search answer comes from pages fetched for that question, with citations. They name different businesses: in our study, only 20.3% of the 2,405 names Gemini gave from memory appeared in Claude's live-search answers for the same industry. In our work, live search is the part a business can change within weeks. Data in memory vs live search.
Search index
A search index is the store of pages a search engine has crawled. Each assistant searches a different one: Google's AI uses Google's, Copilot uses Bing's, ChatGPT uses outside search providers, and Anthropic's subprocessor list names Brave Search, TurboPuffer and Parallel Web Systems for Claude. See Claude SEO.
Retrieval-augmented generation (RAG)
RAG means the system first retrieves documents, then writes its answer from them. The term comes from a 2020 paper by Patrick Lewis and colleagues. It is the general method behind AI answers that cite sources, and the lesson is simple: a page that is not retrieved cannot shape the answer.
Grounding
Grounding is basing an answer on fetched sources instead of memory alone. Google's developer documentation says the model "determines if a Google Search can improve the answer" and, if so, writes searches and answers with citations. Bing's guidelines add that "URLs focused on a primary topic are more likely to be selected for grounding results". The practical rule: one page per topic.
Query fan-out
Query fan-out is how Google's AI Mode handles a question: the model splits it into many related searches and runs them at once. Google's example: for "how to fix a lawn that's full of weeds" it might also search "best herbicides for lawns". Ahrefs found only 38% of pages cited in AI Overviews in early 2026 also ranked in Google's top 10. Our reading: the narrow sub-searches count.
Citation
A citation is a source an AI answer links to. In our study, Claude cited 4,634 sources for 500 buyer questions, 9.3 per answer, from 1,954 different websites, and 1,283 of those sites were cited only once. See which websites AI cites.
Mention vs citation
A mention names your business in the answer. A citation links your page as a source. For at least 36% of the businesses Claude named in English answers (419 of 1,160), their own site was among that answer's sources: 70% in home services, 5.5% in restaurants. Ahrefs found brand mentions correlated with AI Overview visibility at 0.664, against 0.218 for backlinks (a correlation, not proof).
Hallucination
A hallucination is a confident statement that is false. A 2025 paper by Adam Tauman Kalai and colleagues says models "guess when uncertain, producing plausible yet incorrect statements". For a business it means an old price, a wrong address or a mangled name: in our Arabic tests, Gemini labelled Allsopp & Allsopp as المسكن ("the dwelling"). The best defence we know is the same facts everywhere.
Hedged answer
A hedged answer names no business, or only points to maps, reviews or directories. In our study, Claude hedged on 11.6% of Arabic answers against 5.2% in English, and most in regulated fields: 22% of law and 20% of real estate questions. Our reading: there, a firm needs stronger outside proof before AI names it.
Measuring AI visibility
Prompt list (question list)
A prompt list is the fixed set of real buyer questions you check on AI tools, word for word, on a schedule. Answers change from run to run, so a 2026 study by Schulte and colleagues argued that visibility should be measured as a range, not one result. Our free check uses 10 questions; our plans track 25, 40 per language or 60 per language. Try our step-by-step check.
Commercial question
A commercial question is asked by someone close to buying or booking. "Which clinic in JLT does laser hair removal for men?" is commercial; "What is laser hair removal?" is not. Our guarantee counts these: if AI does not name you for at least 3 on your approved list within 30 days, or you are unhappy for any other reason, you get 100% back (rules).
AI visibility score
An AI visibility score is the share of your tracked questions in which AI names you on a given date, split by tool and language: "named in 2 of 10", for example. It only means something with a fixed question list and a dated screenshot behind every result. Be careful with "proprietary scores" that have no published definition.
Share of voice in AI
Share of voice is your share of all business mentions across a question list, next to your competitors. It shows how crowded a field is: in home services, Gemini's five most-named businesses took 46% of its mentions from memory, against 10% for the top five in Claude's live search.
Answer drift
Answer drift is the same question getting a different answer between runs or over time. Asked the same 19 questions about 20 minutes apart, the same Gemini model put the same business first only 4 times. For "Best car service garage in Dubai Marina?" it named Deutsche Technik first, then Royal Swiss Auto Services. One screenshot is a snapshot, not a position.
Model consistency
Model consistency is how far assistants, or versions of one assistant, agree. Not far: on the 421 questions where both named someone, Claude with live search and Gemini from memory shared even one business on 16.4% (25.6% in English, 6.8% in Arabic). On 10 dental questions, two Gemini models agreed on the first name once. Data in do AI assistants agree?
Zero-click search
A zero-click search ends without a visit to any website. Pew found Google users clicked a regular result on 8% of visits with an AI summary, against 15% without one. So we judge AI work by questions won and leads, not traffic alone. More in AI search statistics.
Referral traffic from AI
These are visits that arrive through a link in an AI answer. ChatGPT adds utm_source=chatgpt.com to many links, and on adjoltz.com those visits arrive with the tag and no referrer: 4 in our first two days of tracking (20 and 21 October 2026), one landing on an Arabic blog post and one tapping WhatsApp. Since about 9 October 2026, analysts have seen many ChatGPT links lose the tag, so watch direct traffic too. See tracking leads from AI.
AI Assistant channel
AI Assistant is a default channel group that Google Analytics 4 added in May 2026 for visits from AI tools. Google names ChatGPT, Gemini, DeepSeek, Copilot and Grok as examples, not Claude, and the channel excludes Google's AI Overviews and AI Mode, which count as organic search. A custom channel group can add Claude and Perplexity.
Generative AI performance report
This Search Console report counts impressions of your pages in Google's AI Overviews and AI Mode, by page, country, date and device, and has been live for all sites since 31 August 2026. It shows no queries, so pair it with your own question list to see which questions you win.
Crawlers and site access
AI crawler
An AI crawler is a bot that reads pages for an AI company. There are three kinds: training crawlers (GPTBot, ClaudeBot), search crawlers (OAI-SearchBot, Claude-SearchBot, PerplexityBot) and user-triggered fetchers (ChatGPT-User, Claude-User). Cloudflare measured GPTBot at 30% of AI crawler traffic in May 2025, up from 5% a year earlier. Our AI crawler list covers every major bot.
User agent
A user agent is the name a bot sends when it requests a page, and the token you write after User-agent: in robots.txt. Anyone can claim to be GPTBot, so check the IP address against the lists the companies publish, such as openai.com/searchbot.json, before acting on it.
robots.txt
robots.txt is a text file at your site's root that tells crawlers what they may access. It is a standard (RFC 9309), but its rules "are not a form of access authorization", and a firewall block overrides it. Two traps: a robots.txt that returns a server error means "complete disallow", and a bot with its own named group ignores your User-agent: * rules. Test yours with our AI crawler checker.
GPTBot
GPTBot is OpenAI's training crawler. Blocking it keeps your future pages out of OpenAI's training data, but it does not remove you from ChatGPT search: OpenAI says each setting is independent. Decide on it on purpose, and allow OAI-SearchBot either way.
OAI-SearchBot
OAI-SearchBot is the crawler OpenAI uses to show sites in ChatGPT search. OpenAI says sites that opt out "will not be shown in ChatGPT search answers", and that its systems take about 24 hours to reflect a robots.txt change. For a business that wants to be recommended, this is the OpenAI bot that matters most.
ChatGPT-User
ChatGPT-User fetches a page when a person asks ChatGPT about it, or pastes your link. OpenAI says "robots.txt rules may not apply" to it, because a user asked. Blocking it rarely helps a business: it is the bot that reads your page at the moment a buyer is asking about you.
ClaudeBot
ClaudeBot is Anthropic's training crawler. Anthropic says it collects content that "could potentially contribute" to training, honours robots.txt and supports Crawl-delay. Cloudflare reported that in July 2025 Anthropic's crawlers fetched about 38,000 pages for every visit Claude sent back, the widest gap of the major AI companies.
Claude-SearchBot
Claude-SearchBot "navigates the web to improve search result quality" for Claude users. Anthropic says blocking it may reduce your visibility in Claude's answers, and the same goes for Claude-User, which fetches pages when a person asks a question. Most businesses should allow both. See Claude SEO.
PerplexityBot
PerplexityBot builds Perplexity's search index, and Perplexity says it is "not used to crawl content for AI foundation models". Its user-triggered fetcher, Perplexity-User, "generally ignores robots.txt rules". In August 2025 Cloudflare said Perplexity used undeclared crawlers, and Perplexity disputed it. See our Perplexity guide.
Google-Extended
Google-Extended is a robots.txt token, not a crawler. It controls whether Google may use your content to train Gemini models and to ground answers in Gemini Apps, and Google says it "does not impact a site's inclusion in Google Search". So a block can keep you out of grounded Gemini app answers while you still appear in AI Overviews.
llms.txt
llms.txt is a short Markdown file at yoursite.com/llms.txt that summarises your site for AI agents. Jeremy Howard proposed it on 3 September 2024, and version 2 came in August 2026. Google says Search does not use it, none of the big AI assistants documents reading it before it answers, and SE Ranking found "no correlation" with AI citations across nearly 300,000 domains. We add one as housekeeping. Read our llms.txt guide or use the free generator.
IndexNow
IndexNow is a free protocol that tells search engines which of your pages changed, so they recrawl them sooner. indexnow.org lists Bing, Yandex, Seznam, Naver, Yep, the Internet Archive and Amazonbot; Google is not on the list. Bing's index feeds Copilot, so it is worth turning on. On The Setup Collective, a client site, every deploy sends IndexNow notices. See Copilot SEO.
Bing Webmaster Tools
Bing Webmaster Tools is Microsoft's free console for site owners. Since 10 February 2026 its AI Performance report shows how often your pages were cited in Copilot and Bing's AI summaries, with a sample of the "grounding queries" the AI used.
Server-side rendering
Server-side rendering means the server sends the finished text in the page code, instead of a script building it later in the browser. Vercel found that the crawlers behind ChatGPT and Claude fetch JavaScript files but do not run them, and Google says "not all bots can run JavaScript". A quick test: view your page's source and search for your price. If it is missing, assume AI cannot see it.
Crawl budget
Crawl budget is how many pages a search engine will crawl on your site in a given time. Google says its crawl budget guide is for sites with roughly 1 million or more pages that change weekly, or 10,000 or more that change daily. Most small businesses can ignore it and focus on not blocking bots and keeping key pages linked and in the sitemap.
Entities and structured data
Entity
An entity is one distinct thing a machine can identify: your business, a person, a place or a product. AI has to tie every mention of you, in every spelling, to one entity, and Microsoft asks for "clear and consistent naming for people, organizations, products, and locations". Arabic makes it harder: in our study, Claude wrote 62.2% of business names in Arabic answers in Arabic script only.
Knowledge graph
A knowledge graph is a database of facts about entities and how they connect. Google's Knowledge Graph feeds its knowledge panels. Google says the facts "come from a variety of sources" and that panels appear "when there is enough information available on the open web". Google named the Knowledge Graph among AI Mode's data sources at launch in March 2025.
Schema (structured data)
Schema, or structured data, is a set of labels in your page code that say "this is a business", "this is its phone number", "these are its hours". Google says it uses structured data to understand a page, and that no special schema is needed for its AI features. It makes your facts unambiguous; it is not a switch that gets you named.
JSON-LD
JSON-LD is a format for structured data: a block of code in the page that lists facts as labelled pairs. Google supports three formats (JSON-LD, Microdata and RDFa) and recommends JSON-LD because it is "the easiest solution for website owners to implement and maintain at scale".
FAQPage
FAQPage is the schema type for a list of questions and answers. Google stopped showing FAQ rich results (the drop-down answers under a result) from 7 May 2026, so the markup no longer changes how you look on Google. It still labels questions and answers clearly for any machine reading the page, so our guides keep it.
Organization and LocalBusiness schema
These schema types describe a company (Organization) or a business with a place customers visit (LocalBusiness): name, address, phone, hours, area served and profile links. Google does not show stars for reviews a business controls about itself on these pages, so star markup on your own reviews does nothing in search.
sameAs
sameAs is a schema.org property that links your entity to other pages about it: the "URL of a reference Web page that unambiguously indicates the item's identity", such as a Wikipedia page, a Wikidata entry or an official website. List your main profiles there. Put your Arabic name in alternateName, schema.org's property for another name of the same thing.
Trust signals AI reads
E-E-A-T
E-E-A-T stands for experience, expertise, authoritativeness and trustworthiness, Google's shorthand for reliable content. Google says "trust is most important" and that E-E-A-T "itself isn't a specific ranking factor". In our study, law and real estate, where trust weighs most, were where AI hedged most. Named experts, licence numbers and real cases on your pages are how you show it.
Topical authority
Topical authority is an SEO industry term for a site that covers one subject so fully that search engines and AI treat it as a go-to source. In practice it means one clear page for each real question in your field, linked together. It does not mean mass-produced pages: Google's spam policies call that "scaled content abuse".
"Best of" list
A "best of" list ranks businesses in a category, such as "Top 10 dental clinics in Dubai". AI reads them because they match "best X in Y" questions word for word: in our study, 37.7% of Claude's sources had "best", "top", "leading", "ranked" or "أفضل" in the title or address (65.1% for restaurants), and at least 151 of the 1,747 list-style citations sat on the site of a business the same answer named.
Directories
Directories are listing sites such as Bayut, Property Finder, Practo, Fresha, Clutch or GoodFirms. They were 9.5% of Claude's citations in our study, and big ones reach beyond their field: 176 of Bayut's and Property Finder's 241 citations were for questions that had nothing to do with property. But Practo, TimeOut Dubai, Doctify and ClassPass were not cited once in Arabic answers. Keep every profile complete and identical to your site.
Review signals
Review signals are your ratings, review counts and review text as AI sees them. In our study, review sites were a big source only for restaurants (TripAdvisor, 23.3% of citations), and none of the 4,634 citations was a Google Maps link. Yet 15.2% of answers quoted a star rating or a review count, copied by directories and lists. Google says more reviews and positive ratings help local ranking. See reviews and AI answers.
NAP consistency
NAP stands for name, address and phone. Consistency means those details, plus hours, website and category, are identical on your site, Google Business Profile, directories and social profiles. In our experience, when they differ AI picks the wrong one, hedges or leaves you out. When Adjoltz moved to AI SEO in October 2026, fixing every bio and listing came before any new page.
Language and location
hreflang
hreflang is a tag that tells Google which language or region version of a page to show. Google says "each language version must list itself as well as all other language versions", or the tags are ignored. Google judges a page's language from its visible text, not the lang attribute, so each language needs its own URL. This page and its Arabic twin are linked this way.
Arabic SEO
Arabic SEO is making Arabic pages that search engines and AI can read, rank and cite. Competition is thin: Arabic is the content language of 0.6% of the websites W3Techs tracks, and in the UAE "business setup dubai" gets 9,900 Google searches a month against 140 for تأسيس شركة في دبي. And 90.2% of the citations in our Arabic answers were Arabic pages. See AI SEO in Arabic.
Wrong-country answer (location drift)
A wrong-country answer recommends businesses in another country because the search matched a place name. In our study, at least 11 of 250 Arabic answers described businesses in Saudi Arabia: Al Nahda (النهدة) was confused with An Nahda (النهضة), and Business Bay (الخليج التجاري) with Riyadh's Al Khaleej. In English, a question about an orthodontist in JVC came back with Jacksonville. The fix: write the area, city and emirate in both scripts on every page. Details in Arabic AI answers.
What we'd do for you
The free AI check asks 10 of your buyers' real questions on ChatGPT and Gemini (Google AI) and shows who is named instead of you. The AI visibility audit costs AED 1,950, or USD 549 outside the GCC (adjoltz.com/us), and covers 30 questions per language on four AI tools and a 90-day plan. Monthly plans are Foundation (AED 4,500 or USD 1,249), Growth (AED 8,500 or USD 2,349) and Authority (AED 16,000 or USD 4,399): month to month, no lock-in, no setup fee, with an unconditional 30-day money-back guarantee (rules).
What this glossary doesn't promise
- That knowing the words gets you named. The work is pages, listings, access and consistent facts, checked on a fixed question list.
- That the definitions stay fixed. Bots get added and renamed, and Google's reports and rules change. We date every update.
- That our numbers hold in every market. They come from one run of 500 UAE questions on one day, 21 October 2026.