AI visibility, answered

ChatGPT recommends brands it can find corroborating evidence for at answer time. When browsing is active it runs web searches, reads the top results, and names brands that appear repeatedly across independent, trustworthy sources. Without browsing it falls back on patterns learned in training, which favour brands with a large, consistent published footprint.

Perplexity searches the live web for every query, retrieves a shortlist of pages, and cites the ones whose passages most directly answer the question. It favours pages with a clear, extractable answer near the top, recent publication or update dates, and topical focus — a narrow page that answers precisely often beats a broader, higher-authority one.

AI Overviews run on Google's core ranking and quality systems, then select passages through query fan-out — breaking one question into several sub-queries and retrieving a passage for each. Because selection happens at passage level, most cited pages do not rank in the organic top ten for the original query.

Mostly no. Google renders JavaScript for its own index, but the crawlers used by ChatGPT, Perplexity, and Claude generally fetch raw HTML and do not execute JavaScript. Content that only exists after client-side rendering is invisible to them, even when it is perfectly visible to a human in a browser.

Get your brand named on the independent pages ChatGPT retrieves when it browses: review platforms, comparison articles, industry round-ups, and forum threads. Your own site alone is rarely enough — models corroborate across sources, so a brand described consistently in several independent places is the one that gets named.

Pages that answer one question directly, near the top, under a heading that matches how the question is asked. Comparison tables, specification lists, dated statistics, and short definitional passages are extracted most often, because each can be lifted whole and remain true without the surrounding page.

Indirectly, yes. No major AI engine has confirmed that structured data is a ranking input, but schema makes a page's facts unambiguous — what it is, who published it, what it is about, and which entities it refers to. That resolves the reconciliation problem engines face, and unambiguous pages get misread less often.

Barely, on current evidence. Around 9–10% of major sites publish an llms.txt, but monitoring of AI crawler traffic shows the file is almost never requested — GPTBot, ClaudeBot, PerplexityBot and Google-Extended overwhelmingly crawl HTML directly. No major AI company has committed to reading it, and Google has said it does not support it.

Indirectly. AI engines do not appear to use links as a direct citation input, but the engines that browse retrieve from search results, and links still influence what ranks. More importantly, the behaviour that earns links — being referenced by independent sources — is the same behaviour that gets you named in AI answers.

Yes, substantially. Reddit threads are retrieved and cited heavily by AI engines because they contain candid, comparative, first-hand opinion that is hard to find elsewhere. A recommendation thread in a relevant subreddit frequently ends up as a cited source behind a brand recommendation.

They are among the strongest single signals. AI engines routinely cite review platforms when justifying a recommendation, and a brand with no verifiable review presence gives the model nothing to point at. Engines will sometimes state outright that they could not find reviews for a brand — and then recommend a competitor whose ratings they can quote.

It helps, but it is a consequence rather than a tactic. Wikipedia and Wikidata are core reference points for the knowledge graphs AI engines reconcile entities against, so a well-sourced entry strengthens recognition. Creating one for a brand that does not meet notability standards will fail and can damage your standing.

Yes, more than in traditional search. Many questions carry an implicit 'right now', and engines resolve that by preferring pages with visible, recent dates. Undated content is systematically disadvantaged — an engine cannot tell whether it is current, so it reaches for something it can date.

Ask the engines the questions your buyers ask, then record whether you were named, in what position, who was named instead, and which sources were cited. Run the same prompt set on a fixed schedule so results are comparable. Ad-hoc spot checks feel informative but cannot show movement.

Monthly for most brands. AI answers fluctuate run to run, so measuring more often mostly captures noise, while measuring less often means a competitor can take a category for a quarter before you notice. Move to fortnightly during an active push, and after any significant site or positioning change.

Count how many of your prompts named each brand, then express yours as a share of all brand mentions across the set. If forty prompts produce two hundred brand mentions and you account for twenty, your share of voice is ten percent. Weight by position if you want prominence reflected as well.

On a 0–100 scale, most established brands land between 40 and 65. Above 70 usually means an engine names you unprompted in open category questions; below 35 means you are found only when someone already knows your name. The number matters far less than which underlying signals are dragging it down.

Query fan-out is when an AI engine breaks a single question into several related sub-queries, searches each one separately, and synthesises the results into one answer. Asking for 'the best CRM for a small agency' might fan out into searches for best CRM 2026, CRM pricing for small teams, and agency CRM reviews.

Traditional search returns a ranked list of links and leaves the judgement to you. AI search returns one synthesised answer with a handful of citations, having already made the judgement. The competition shifts from ranking above other pages to being one of the few sources the answer is built from.

Answer Engine Optimization is the practice of shaping content and brand signals so AI answer engines surface, cite, and recommend you inside their responses. Where SEO competes for a position in a ranked list of links, AEO competes to be part of the single synthesised answer the engine returns.

Generative Engine Optimization is the practice of earning citations and mentions inside generative AI responses. In practice GEO and AEO describe the same work with different emphasis — GEO stresses being cited by generative systems, AEO stresses being the answer. Most practitioners use the terms interchangeably and neither is a distinct discipline.

Entity SEO is making sure search and AI systems understand your brand as a specific, identifiable thing rather than a string of characters. It matters because AI engines reconcile what they read against a knowledge graph — a brand they cannot resolve confidently is one they are reluctant to name.

Yes — arguably more, in a narrower way. Google's AI features run on the same core ranking and quality systems as search, and the engines that browse retrieve from search results. If you are not crawlable, indexable, and credible, you are not in the pool an AI answer is assembled from.

Usually one of four reasons: your site blocks AI crawlers, your content only renders in JavaScript, nothing independent has been published about you, or your brand name is ambiguous enough that the model cannot resolve which company you are. The first two are technical and fixable in a day.

Almost always because the wrong information is findable and the right information is not — an outdated price on a directory, a stale description on a profile you forgot, or a competitor's page describing you inaccurately. Models reconcile conflicting sources by favouring what appears most consistently, not what is most authoritative.

Because the engine can find more, and more specific, independent evidence about them. Typically they hold review profiles you lack, appear in comparison articles you are absent from, or state concrete claims — pricing, audience fit, certifications — that give a model something quotable. Recommendation follows evidence, not product quality.

Decide the canonical version of every core fact, publish it plainly on your own site in both text and structured data, then propagate it to every profile you control and request correction everywhere you do not. Inconsistency is penalised because a model that finds three answers will confidently state none of them.

Not if you want to appear in AI answers. Blocking GPTBot, ClaudeBot, PerplexityBot, or Google-Extended removes your pages from the retrieval pool those engines draw on, which removes you from the answers. Blocking is a defensible choice for publishers protecting licensable archives, and a self-inflicted wound for almost everyone else.

Technical fixes such as unblocking crawlers or server-rendering content can show up within days on browsing engines. Content and structural work typically moves the needle in four to eight weeks. Earning independent coverage and reviews — the strongest signals — usually takes one to two quarters to compound.

Retrievability, independent corroboration, extractable answers, entity clarity, and freshness. In rough order of leverage: be crawlable and server-rendered, be described consistently by sources other than yourself, answer specific questions directly near the top of a page, resolve unambiguously as an entity, and show recent dates.

Only if it adds coverage of questions you did not previously answer. Volume alone does nothing — an engine retrieves one passage per sub-query, so a second article on a question you already answer well competes with yourself. Breadth across a topic's fan-out beats depth on any single point.

Not in the organic answer itself. Some engines run clearly labelled ad units alongside generated responses, but the cited sources behind an answer are selected by retrieval and cannot be bought. Any vendor offering guaranteed placement inside AI recommendations is describing something that does not exist.

Treat it as passage-level SEO. AI Mode fans a query out into sub-queries and assembles an answer from passages, so the goal is clean, self-contained answers to each sub-question rather than one page ranking for the head term. Most pages cited in AI Mode do not rank in the organic top ten.

By being present on the software review platforms and comparison content that AI engines cite for software questions, with an explicit statement of who the product is best for. B2B buying questions are highly constrained — team size, budget, compliance, integrations — and engines recommend whichever product matches the stated constraint.

Through consistent local citations and reviews. AI engines answering location questions lean on map profiles, directory listings, and review platforms, and they need your name, address, and phone number to match exactly across all of them. Mismatched details are the most common reason a local business is skipped.