Controlled access
Robots, canonicals, internal links, HTTP status codes and sitemaps should let authorised systems reach decisive pages.
REFERENCE / DISCOVERY
AI and search ecosystem
A practical map of the interfaces that discover, rank, synthesise and cite information — and the disciplines needed to build a credible presence across them.
00 / FIELD MAP
Visibility no longer depends on one engine
The same question may start in conventional search, continue in an AI assistant, trigger a comparison and end on an evidence page. Strategy must treat indexing, experience, entities, sources and measurement as one chain.
12AI assistants12search engines6disciplines
REFERENCE / 2026
Four signals to consolidate
Conventional and generative engines do not require another magic layer. They need reliable access, distinctive content, readable evidence and regular observation.
Original answer
Each page should resolve one clear intent with first-hand information, a clean structure and value beyond paraphrasing other sources.
Attributable evidence
Author, entity, dates, sources, images and structured data should clarify who claims what, on which basis and in which context.
Continuous observation
Rankings are no longer enough: track citations, reused pages, factual errors, real journeys and variations between interfaces.
PLATFORMS / 01
Major public AI assistants
These interfaces do not answer in the same way. Some favour conversation, others sourced research, reasoning, code or work-environment integration. Good practice is to observe several surfaces instead of optimising for a single brand.
01General-purpose assistant
ChatGPT
Assisted research, writing, analysis, code and multimodal work in a conversational interface.
Check Open cited links and distinguish a web-supported answer from one with no explicit source.
02Reasoning & production
Claude
Document analysis, reasoning, structured writing and software development with extended context.
Check Check that retrieved document passages cover the whole question and that the document date remains relevant.
03Multimodal assistant
Gemini
Google assistant for research and production across text, images and its service ecosystem.
Check Compare the summary with displayed source pages and note claims absent from those pages.
04Assistant & search
Microsoft Copilot
Conversational interface connected to Bing search and, depending on the plan, Microsoft work environments.
Check Identify whether the answer uses the public web or connected work data in the current setting.
05Answer engine
Perplexity
Conversational web research with synthetic answers, visible citations and source exploration.
Check Read cited pages: a link does not ensure a faithful summary or a reliable source.
06Work assistant & agent
Mistral Vibe
Vibe, formerly Le Chat, brings together Mistral AI chat, productivity and coding uses. Identify the selected mode before comparing its results.
Check Check whether web search is enabled and whether current facts link to accessible sources.
07Reasoning & code
DeepSeek
Assistant focused on reasoning, programming and technical problem-solving through a web interface.
Check Recheck recent and technical facts against primary sources before making a decision.
08Consumer assistant
Meta AI
Assistant integrated into the Meta ecosystem for questions, creation and multimodal interactions.
Check Check sources and distribution context when an answer circulates inside a social app.
09Multimodal assistant
Grok
SpaceXAI web assistant for conversation, research and creation across multiple content formats.
Check Open the sources for a recent fact and distinguish a web-grounded answer from a model-only response.
10Multi-model platform
Poe
An interface for official and third-party bots; provider and rules vary with the selected bot.
Check Record the selected bot and its provider before comparing answers: Poe is an interface, not one model.
11Multimodal assistant
Qwen
Assistant in the Qwen ecosystem for conversation and work with text and multimodal content.
Check Check whether web search ran and compare current claims with the pages offered as sources.
12Research & documents
Kimi
Moonshot AI assistant combining conversation, connected search and document work.
Check Compare cited passages with the full document and note context lost in long summaries.
INDEX / 02
Major search engines
Search engines differ by index, providers, privacy policies, markets and answer formats. Robust visibility covers structural engines without ignoring alternatives relevant to the audience.
01General web index
General search engine with web, local, image, video and enriched answer surfaces across many markets.
Check Compare canonical URL, index status and actual display of important pages in Search Console.
02Index & Microsoft services
Microsoft Bing
Microsoft web search, also important through its data, webmaster tools and AI integrations.
Check Check crawling and indexing in Bing Webmaster Tools before interpreting AI citations.
03Privacy-focused search
DuckDuckGo
Search designed to limit profiling, with instant answers and optional AI features.
Check Test important queries and open results instead of inferring coverage from the home page.
04Independent index
Brave Search
Private search based on its own index, also offering synthetic answers with supporting sources.
Check Compare pages present in the index and verify sources cited in synthetic answers.
05European search engine
Qwant
French privacy-oriented search engine with web search and AI-assisted answers.
Check Test searches in target markets and check the origin of AI-assisted answers.
06Impact-led search
Ecosia
European search engine using revenue for climate action and combining several result providers.
Check Observe results actually shown for target queries separately from the environmental promise.
07Private search
Startpage
Search intermediary designed to provide results without building a personal user profile.
Check Distinguish the private intermediary from the coverage of the index supplying results.
08Portal & search
Yahoo Search
Long-standing search engine integrated with a content, news, finance and services portal.
Check Test target pages and queries in the portal rather than assuming presence from another engine.
09Subscription search
Kagi Search
Ad-free search with customizable results and a subscription-based offering.
Check Test priority queries with the same personalization settings and record suppressed or boosted results.
10Independent index
Mojeek
Search engine with its own crawler and index, designed without individual search tracking.
Check Check important pages in this independent index before inferring presence from Google or Bing.
11Multiregional search
Yandex Search
Search engine whose results can be examined by language, region and audience query.
Check Compare results by region and language for queries relevant to the actual target market.
12Chinese-language search
Baidu
Chinese-language search engine and portal; assess coverage with queries and pages relevant to that market.
Check Test relevant Chinese queries and inspect displayed pages rather than inferring coverage from another market.
24 rubrics to document your comparisons
Choose the same question, scope and authorized records. Retain every answer, then classify checks separately as supported, limited, contradicted or unobserved. Do not average away an inaccessible reference or uncertain identity.
Proposed method, with no service score or measured result. In the JSON, observations, test dates and conclusions are left empty for your own log.
PANEL / OBSERVATION
Build a panel that supports a decision
The directory does not require monitoring every platform. Select those relevant to a real audience and keep a comparable method across them.
Define the audience
Specify countries, languages, topics and tasks. Include a platform only when these people can realistically find or verify information there.
Separate the surfaces
Record whether you are observing an index, a result aggregator, an assistant that can search or a multi-bot platform. These roles do not produce equivalent evidence.
Fix the tests
Keep a dated set of queries and questions, languages, settings and reference URLs. Open results and citations; record accuracy, source and entity confusion.
Set a cadence
Repeat observations at defined intervals and after major changes. Address verifiable errors and useful journeys first; a single screenshot is not a trend.
READING / RESULTS
Interpret what the panel shows
A presence, absence or isolated citation does not identify its cause. Each case links a finding to a check and a bounded decision.
Indexed page, never cited
Check : Compare tested questions, competing pages, the visible answer and access to the primary source. Indexing alone does not measure assistant selection.
Reasonable decision · Improve the answer and provenance if a real need is unmet; keep observed frequency as a measure without promising citation.Name mentioned, wrong entity
Check : Record exact wording, namesakes, dates, identifiers and links associated with the answer.
Reasonable decision · Clarify ambiguous reference pages and controlled profiles, then monitor question variants after publication.Link cited, claim unsupported
Check : Open the URL and find the passage meant to support each claim. Note when the link only covers the broad topic.
Reasonable decision · Log the synthesis error, clarify an ambiguous source and avoid reusing the answer as evidence.Different results by language
Check : Compare translated intent, canonical pages, available sources and test settings.
Reasonable decision · Fix missing language or context in the corpus, then repeat the same panel in each target language.More citations, stable contacts
Check : Separate mentions, clicks, qualified visits and actions; check volumes and dates before concluding.
Reasonable decision · Keep both series and examine the journey to a decision; do not attribute business impact to citation alone.METHODS / 03
SEO, SXO, AEO, GEO and LLMO
These acronyms are not silos. They describe stages of one journey: be discovered, understood, provide the right answer, earn citation and deliver a useful experience.
SEO
Search Engine Optimization
Make pages crawlable, indexable, understandable and useful enough to earn durable organic visibility.
Technical · content · authorityExplore the methodSXOSearch Experience Optimization
Connect the promise shown in search results to a fast, readable, reassuring page that supports the right action.
Intent · experience · conversionExplore the methodAEOAnswer Engine Optimization
Formulate clear answers to real questions while retaining the context and evidence needed for verification.
Questions · answers · snippetsExplore the methodGEOGenerative Engine Optimization
Structure information so generative engines can identify the entity, synthesise without distortion and cite the right sources.
Entities · sources · citationsExplore the methodLLMOLarge Language Model Optimization
Strengthen the public coherence of an entity, its relationships and corpus to improve retrieval, representation and attribution.
Corpus · graphs · coherenceExplore the methodMESUREVisibility observability
Track rankings, clicks, journeys, citations, reused sources, answer accuracy and share of presence across a stable panel.
SERP · analytics · AI answersExplore the methodCHAIN / 04
From crawling to citation
A durable presence follows a logical order. Each stage depends on the previous one and has its own controls.
- 01
Crawl
Robots and retrieval systems must reach pages, follow links and identify canonical versions.
- 02
Index
Important content must be technically eligible, distinctive, fast and associated with the correct index status.
- 03
Understand
Headings, structure, data, authors, dates, entities and relationships create a coherent representation.
- 04
Select & cite
A useful, original and evidenced answer is more likely to be retained as a source or recommendation.
- 05
Serve & measure
The page must fulfil its promise, guide the user and create observable signals without manipulating the journey.
OBSERVATORY / 05
What actually needs measuring
No single metric proves visibility. A useful dashboard connects exposure, understanding, experience and outcome.
Coverage
Crawled, indexed, excluded and canonicalised pages, plus freshness of decisive content.
Demand
Queries, questions, intents, seasonality and changes in audience language.
Organic presence
Rankings, impressions, clicks, click-through rate and visibility share by intent group.
Experience
Speed, accessibility, comprehension, reading depth, navigation and task success.
AI presence
Observed answers, mentioned brands, cited sources, accuracy and stability across models.
Impact
Qualified contacts, supported decisions, trust, conversion and reduced representation errors.
CONTROL / 2026 MATRIX
Eight checks before publishing
A page ready for search and AI assistants must remain useful to people, technically accessible and verifiable over time. This matrix turns those principles into practical gates.
Allowed crawling
Check robots.txt, meta directives, HTTP status, click depth and internal links.
Expected evidence · Tested URL, 200 response and internal pathReference version
Align canonical, redirects, hreflang, parameters and language variants.
Expected evidence · Absolute canonical and reciprocal hreflangIdentifiable answer
Give each page one central question, an explicit title and an immediately findable summary.
Expected evidence · Single intent and standalone answerCoherent entities
Name people, organisations, places, offers and relationships consistently across the site.
Expected evidence · Matching names, profiles and relationshipsReadable provenance
Tie important claims to an author, date, method or accessible source.
Expected evidence · Visible author, date and sourceUsable images
Use distinct, sized, lightweight and accurately described visuals attached to their page.
Expected evidence · Dimensions, alt, weight and metadataVerified experience
Check mobile, keyboard, contrast, speed, filters, forms and overflow.
Expected evidence · Real journey without blocking errorDated observation
Keep a query panel and record mentions, citations, source pages, errors and conversions.
Expected evidence · Comparable log over timeSOURCES / 06
Official tools and documentation
Interfaces and rules change. Technical decisions should return to engine documentation, webmaster tools and open standards.
01Open resource02Open resource03Open resource04Open resource05Open resource06Open resource07Open resource08Open resource09Open resource10Open resource11Open resource12Open resource13Open resource14Open resource15Open resource16Open resource17Open resource18Open resource19Open resource
Google Search Central
Crawling, indexing, structured data and Google Search best practices.
Google Search Console
URL inspection, performance, coverage and technical site signals.
Bing Webmaster Tools
Search reports, inspection, backlinks and Bing indexing tools.
IndexNow
Protocol notifying participating engines of added, changed or deleted URLs.
Schema.org
Shared vocabulary for describing entities, content, organisations and relationships.
Web.dev
References for performance, accessibility, quality and web experience.
Google Search · IA
Official principles for making a site eligible for Google Search AI features without artificial GEO shortcuts.
OpenAI · Publishers and developers FAQ
OAI-SearchBot access controls and how content may appear in ChatGPT search.
Bing · AI Performance
Bing Webmaster Tools reporting for citations and source pages observed in AI answers.
Google Images · SEO guide
Official guidance for images, alternative text, dimensions and image sitemaps.
xAI · Grok overview
Official overview of Grok and its access points.
Poe · FAQ
Official help on bots, their providers and how Poe works.
Qwen · Qwen Chat
Official overview of the Qwen interface and supported formats.
Kimi · User guide
Official guide to Kimi and its research functions.
Kagi · Search help
Official documentation for the search engine and its settings.
Mojeek
Official explanation of Mojeek’s independent index and crawler.
Yandex · Search help
Official help on queries, regions and site-owner information.
Baidu · Search
Official interface for observing results of relevant Chinese queries.
Mistral · Vibe (formerly Le Chat)
Official rename announcement distinguishing Work, Code and Chat modes.
07 / FAQ
Frequently asked questions
Essential distinctions to avoid shortcuts and false promises.
Should every AI be optimised separately?
No. A shared base — original content, clear entities, strong sources, technical accessibility and coherent reputation — serves several systems. Differences are then measured by platform.
Does GEO replace SEO?
No. Assistants and generative engines still rely on discoverable, retrievable and credible content. GEO extends SEO towards synthesis, attribution and citation.
Can an AI citation be guaranteed?
No. No markup or file guarantees citation. Quality, clarity, provenance and authority can only improve the available evidence.
Must a site appear in every search engine?
Cover the surfaces used by target audiences. Google and Bing often structure much of the field, while private, European or specialised engines may matter by market.
Do these logos imply a partnership?
No. They only identify platforms in this independent directory. No partnership or commercial recommendation is implied.
How is AI visibility tracked?
Use a stable question panel, dated observations, mentions, citations, errors, variations and comparison with organic and behavioural data.
08 / CONTACT
Turn this directory into an action plan
An audit connects the platforms that matter, priority intents, technical state, evidence corpus and measurement protocol.
DEEPER READING / PRACTICE
Compare surfaces according to a specific task
Visibility observations are more useful when they distinguish presence, relevance and accuracy. Every result should remain linked to a question, a platform and the known test conditions.
Editorial supplement added on .
Starting situation
The reader is looking for a tool to find a source, explore a topic, or get a summary. Surfaces should be compared according to the task, the results available and the possibility of checking the answers.
Priority checks
Run the same documented question on the selected surfaces. Observing links, access to documents, accuracy of the context and ease of resuming the search from the result.
Interpretation pitfalls
A longer or faster response is not necessarily more useful. The test conditions and the type of question influence the comparison.
Working record to retain
A observation sheet indicates task, query, surface, result and limit. The final choice is based on the needs of the route rather than on a universal ranking.
Apply the approach to an initial case
Preparing a small series of questions representing the needs of the public. Keep their formulation and define the categories before the test. An unexpected observation becomes a case to be examined: check the source, context and conditions, and then decide if a repeat or other question can clarify the result.
Present the result and its limitations
Presenting the raw results next to the comment allows you to check the rankings selected. A comparison should specify what is common between the observations and what differs. When the conditions are not close enough, keep the series separate and explain the need for a complementary collection.
PRACTICAL CASE / WORKED EXAMPLE
The best tool depends on the research stage.
This pedagogical case shows how to qualify a research result or a generated response. It distinguishes the punctual observation, its verification and the conclusion that the test conditions actually support.
Example context
A reader wants to find a precise publication and then explore its references. One surface gives a clear summary, while another makes it easier to continue to source documents.
Suggested approach
Take both steps and find out the results. Compare identification, access, context and continuation. Examine the useful passages rather than just remember the quality of presentation of the answer.
Conclusion and next step
The path can combine several surfaces with a separate role. The choice is motivated by the observed tasks and the possibilities of verification available.
Checklist for applying this example
- Keep the formulation of the question and the conditions known.
- Separate presence, relevance, attribution and consistency with the source.
- Notes the non-exploitable cases and their effect on the denominator.
- Compare compatible series and keep the original observations.
Questions for deeper practice
Is it possible to use multiple surfaces in one search?
Yes, keeping the origin of the results and the role of each step. The documents found must then be examined directly to verify the proposals useful to the file.
How to distinguish a useful observation from a too general conclusion?
Name the question, surface and conditions of the test. The observation describes what has been seen; the conclusion must explain to which whole it can apply. When the cases are few or selected for a particular purpose, the presentation of the results retains that scope instead of presenting a general property of the platform.
Why keep the full answers and sources displayed?
An isolated capture or a rate incompletely summarizes the result. The complete answer allows you to review the proposals, their reservations and related links. These elements facilitate a second reading, a subsequent comparison and the understanding of a ranking, especially when a quotation is present but supports only a part of the statement.
DETAILED WORKSHOP / APPLICATION
Composing a research according to its successive steps
This workshop offers a five-step sequence to apply the page labels. The entries, observations and conclusions are kept together in order to facilitate the verification and recovery of the work.
01 / Prepare the necessary information
Define the need: find a document, explore a topic or compare information. Choose surfaces to test and a common question.
02 / Carry out the approach
Exercise the research and explore the possibilities of continuation. Examine access to sources, context and ease of identifying useful parts. Compare the results according to the expected role of each step.
03 / Examine an alternative situation
If a surface is well oriented but does not give access to the room, continue with another entrance point. Keep the source to understand the contribution of each consultation.
04 / Decide on the next step
The path is useful when the decisive documents can be examined and linked to the issue. The complementarity of the surfaces should be observed rather than supposed.
05 / Hand over the verification records
Transmit requests, surfaces, results and selected documents. The steps can be repeated without confusing synthetic response and final documentary evidence.
ANNOTATED WORKING DOCUMENT / REVIEW GRID
Role of Surface Research
The work document notes task, request, surface, result and continuation. It shows which consultations guide the research and which parts allow to verify the information selected.
On a small screen, scroll the grid horizontally to read all three columns.
| Observation | Supported interpretation | Next step |
|---|---|---|
| The surface recovers the room and makes it easy to read. | It responds to this stage of need in the conditions observed. | Keep the track and check the useful document directly. |
| The synthesis is oriented but does not show the decisive source. | The result helps to explore without still providing sufficient evidence. | Follow by references or another search area. |
| The ranking is based only on the form of the answer. | The assessment does not reflect the origin and verification requirements. | Go back to the task criteria before choosing a use. |
Reviewing the working document
The final verification compares contributions and possibilities of continuation. Several surfaces can be complementary if their results remain linked to the documents and questions that structure the file.
REPORTING / EXAMPLE WORDING
Presentation of results: Compare surfaces according to a specific task
Point to clarify
A surface can be useful in one step without meeting all the research needs.
Before handing over
Present the role of each surface and the continuation actually performed.
Example of a qualified conclusion
The consultation brought the guidance or the detail described. The documents used for verification are examined directly and kept with their origin.