AI visibility news, insights and playbooks
- Vizible AI MCP connector: ask Claude or ChatGPT about your AI visibility
Every Vizible AI account, brand or agency, can now connect its data to Claude, Claude Code, ChatGPT or Mistral's Vibe through one MCP address. The assistant reads your visibility, competitors, cited sources, tasks and AI traffic, and never changes anything.
- No private equity firm is named in more than 15% of AI chatbot answers
A 4,000-answer study of ChatGPT, Gemini, Perplexity and Grok found that even KKR, the most visible name in private equity, gets named in only 14.6% of relevant AI answers. Across 98 ranked firms the average sits at 3%, and data platforms like PitchBook and Preqin are cited more often than any single firm's own website.
- Nearly 3 of the top 5 AI-recommended brands change once you leave New York and English
Nearly three of the top five brands ChatGPT and Gemini recommend change once a buyer moves away from New York and English, according to a September 2026 study of 1,200 identical purchase questions. The pattern is sharpest in payments and HR software, and the two AI models often don't even agree with each other.
- Vizible AI for agencies: every feature to track and report AI visibility for your clients
Vizible AI gives agencies a dedicated workspace for every client, daily tracking on ChatGPT, Gemini, Perplexity, Copilot and Google's AI, and white-label reports to share. Here is every agency feature, from client workspaces to weekly recommendations.
- Vizible AI for brands: every feature to track and grow your visibility in AI answers
Vizible AI asks six AI platforms the questions your buyers ask, every day, and shows where your brand is named, which sources AI engines cite and what to do next. Here is every feature for brands, from AI Share of Voice tracking to the website audit and daily tasks.
- On Perplexity, 64% of retrieved pages are never cited. Retrieval and citation measure different things.
On Perplexity, 64 percent of the pages it retrieves for a query never make it into the cited answer, according to a February 2026 study of over a million AI citations. Two more 2026 papers map exactly where the rest of that funnel breaks down, and why clearing it on one AI engine says nothing about another.
- Brands with active review profiles get cited by AI 75% of the time. Brands with none get 1%.
A brand with no claimed review profile gets cited by AI chatbots about 1% of the time. One with an active, well-reviewed profile gets cited more than 75% of the time, according to a 2026 study of over 800,000 AI responses. Here's what actually drives the jump, and why faking it is a bad idea.
- ChatGPT's in-chat checkout converted three times worse than Walmart's site. OpenAI moved on in March.
Walmart said purchases inside ChatGPT converted three times worse than clicks to its own site, and OpenAI confirmed in March 2026 it's scaling back Instant Checkout. The product feed and eligibility rules that decide whether a brand shows up at all haven't gone anywhere.
- Cloudflare added an AI opt-out flag to robots.txt. Google says no crawler obeys it.
Cloudflare's Content Signals Policy added three new preferences to robots.txt for controlling AI crawlers. Google says the signal does nothing, while Anthropic's three-way crawler split shows a more verifiable alternative already exists.
- ChatGPT knows 93% of local businesses. It still won't recommend four in ten of them.
A 10,000-business study of ChatGPT and Perplexity found that AI engines know most local businesses but recommend fewer than half of them. Google reviews, not technical SEO, turned out to be the strongest lever separating the recommended from the invisible.
- Adding quotations lifted AI citation scores 41% in the study that started GEO
A 2024 Princeton, Georgia Tech, and IIT Delhi study tested nine ways to rewrite content for AI answer engines, and only four moved the needle by double digits. A year of follow-up research mostly backs the pattern up, but not every number travels unchanged to 2026.
- ChatGPT's GPT-6 now builds calculators and charts in answers. OpenAI never explained how sources get credited.
OpenAI's GPT-6 update lets ChatGPT answer with interactive charts, calculators, and comparisons instead of plain text, and the company says GPT-6 Instant responds to web-search questions 44% faster than its predecessor. What its announcement never explains is how, or whether, these generated interfaces credit the sources behind them, a gap that matters for anyone tracking AI visibility.
- Hreflang tells Google which language page to show. Most AI engines ignore it.
Hreflang tells Google which language version of a page to serve. New testing and two 2026 research papers show ChatGPT, Claude, and Perplexity mostly skip that signal entirely, defaulting to English sources even when a local-language page exists. Here's what their own crawler documentation says, and what to build instead.
- ChatGPT's Deep Research reads your pages closely. Most of it never earns a citation.
ChatGPT's Deep Research can open dozens of pages about a brand in a single session and still cite almost none of them. New log analysis, subscription-tier data, and an academic citation-accuracy study together explain why reading and citing are two separate decisions, and what a brand can do about the gap.
- Information gain was supposed to reward unique content. Google's patent describes something narrower.
Google's "information gain" patent has become SEO shorthand for "unique content ranks better," but a close reading shows it compares a page to what one user already saw, not to competitors. Independent 2026 research on AI citations tested the uniqueness claim directly, and found something narrower but still useful.
- AI crawler traffic tripled after GPT-5. Rate-limiting the wrong bot can quietly cost you citations
AI crawler traffic has roughly tripled since GPT-5 launched, and some sites are throttling or blocking bots outright to cope with the server strain. The problem is that OpenAI's own documentation shows GPTBot, OAI-SearchBot, and ChatGPT-User do very different jobs, so a blanket rate limit can choke off the citations a site was actually trying to protect.
- How to get cited by Microsoft Copilot, the assistant running three different AI models at once
Microsoft Copilot runs on a shifting mix of OpenAI, Anthropic's Claude, and Microsoft's own in-house MAI models, all grounded through Bing's search index. Most AI visibility tools still track it as one static assistant. Here's what Microsoft's first real citation data, published in February 2026, actually shows about what gets cited and why.
- Well-known brands win nearly every AI recommendation, until a slightly better rating shows up
Three AI models recommended the well-known brand over an unfamiliar one 100% of the time in a 2026 study, until a rating edge of less than 0.1 stars broke the pattern completely. A separate experiment found the same bias is already shaping real purchase decisions outside the lab.
- Canonical tags tell Google what to index. AI crawlers never confirm they're listening.
OpenAI, Anthropic, and Perplexity's crawler documentation never mentions canonical tags, the signal that tells Google which duplicate page to index. Here's what GPTBot, ClaudeBot, and PerplexityBot actually use to pick between near-identical pages instead, and whether the tag is still worth keeping.
- Chosen brands appeared twice as often in ChatGPT's answers, and branded AI Share of Voice missed it entirely
Branded AI Share of Voice checks whether an AI engine recognizes your company. Non-branded AI Share of Voice checks whether it recommends you without being told to, and a 2026 Profound shopping study shows that's the number tied to what buyers actually pick, not the one in most dashboards.
- Query fan-out splits every search into dozens of sub-queries. Most of them don't affect who gets cited.
Query fan-out splits every AI search into dozens of background sub-queries, and the GEO industry built a whole playbook around "covering" them. An 800,000-page study tested that playbook against real citation data in April 2026, and it didn't hold up. Retrieval rank and relevance mattered far more than coverage ever did.
- Why the same brand ranks completely differently on every AI engine
BrightEdge's analysis of five AI engines found they agree on which brands to recommend only 36% to 55% of the time, and the overlap drops to 60% in healthcare and lower still for individual brands. A company can rank in the top 10 on one engine and disappear from the top 100 on another, which makes single-engine AI visibility tracking actively misleading.
- Mistral renamed Le Chat to Vibe, and most AI visibility tracking hasn't caught up
Mistral renamed Le Chat to Vibe in May 2026, then raised 3 billion euros at a 21 billion euro valuation four months later, yet most AI Share of Voice trackers are still watching for the old name. Here's how Vibe's crawlers, web search, and citations actually work, and why the platform barely registers in AI visibility reports today.
- Noindex doesn't block AI crawlers. Robots.txt does, but not reliably
Most sites assume a noindex tag keeps AI chatbots from citing a page. OpenAI, Anthropic, and Perplexity's crawler documentation never mentions meta tags at all, only robots.txt, and 2026 data shows even that isn't always honored.
- AI assistants judge unfamiliar brands by what their name literally means
When an AI model has no real information about a company, it quietly fills the gap by reading the words in the brand's name, according to new research from Peec AI. The bias vanishes for household names like Slack and Discord, but it's a measurable risk for new and category-creating brands AI engines haven't learned yet.
- AI assistants send visitors to 404 pages 2.87 times more often than Google, and ChatGPT is the worst offender
AI assistants send visitors to 404 pages 2.87 times more often than Google, and ChatGPT is the biggest offender by a wide margin. Ahrefs, SE Ranking, and crawler-log analysis from Urllo all point to the same cause: stale indexes, undocumented redirect handling, and a crawl budget that wastes over a third of its requests on dead pages. Here's what the data shows, and the redirect hygiene that actually fixes it.
- AI engines cite content 25.7% fresher than Google, and faking a publish date doesn't fool them for long
Ahrefs analyzed 17 million AI citations and found engines cite content 25.7% fresher than what shows up in organic Google results. A separate 2025 academic study found large language models can be fooled by a fake publish date, but only briefly, and only up to a point.
- How to fix wrong AI answers about your brand, platform by platform
ChatGPT, Google, Gemini, Perplexity, and Claude each let you report a wrong answer about your brand, but none of them promise it gets corrected. Here's what each platform's process actually does, and the source-level fix that works when the report button doesn't.
- How to get cited by Meta AI, the billion-user assistant your GEO tracking probably ignores
Meta AI has more than a billion monthly users but just 0.1% of the dedicated chatbot market, and Meta-WebIndexer, the crawler behind its citations, saw its share of AI crawler traffic jump seventeen-fold in a single month in 2026. Here's how Meta's three crawlers work, and how to check whether your site can actually be cited.
- ChatGPT's AI prompt share dropped 20 points in 2026, and most AI Share of Voice tracking hasn't caught up
ChatGPT's share of US AI prompts fell from 70% to 50% between January and June 2026, while Gemini and Claude gained sharply, according to Comscore. Here's what that shift means for how brands should weight their AI Share of Voice tracking across engines.
- Topical authority beats domain rank for AI citations, new research shows
Two independent 2026 studies agree: how much of a topic your site covers predicts AI citations far better than domain authority does. One found a 220x citation gap between wide and narrow topical coverage. Here's what the data actually shows, and how to build the coverage that gets rewarded.
- Fake AI crawlers hit 824 IP addresses in 2026, and most sites still can't tell the real ones
GreyNoise found 824 IP addresses impersonating GPTBot, ClaudeBot, and PerplexityBot to hunt for exposed credentials, and none of them matched the real companies' published ranges. Here's how OpenAI, Perplexity, Anthropic, and Cloudflare each let (or don't let) you verify which AI crawlers hitting your site are genuine.
- Press releases and AI citations: what a 99% stat actually measures
A widely shared PR-wire statistic claims 99% of press releases get cited by ChatGPT and Claude. Independent analyses of millions of real AI citations tell a different story: press releases barely register, and owned newsroom pages beat wire distribution by a wide margin. Here's what each study actually measured, and what it means for brand teams tracking AI visibility.
- Blocking AI training can quietly block your AI citations too
Cloudflare's Content-Signal now lets a site separate three different AI permissions in one robots.txt line: search indexing, real-time AI input, and model training. Most sites still set one blanket rule for all three, which means blocking AI training often blocks the citations a brand actually wants from ChatGPT and Perplexity.
- Siri AI launched running on Google Gemini, and Apple won't say how it picks sources
Siri AI launched on September 14, 2026, running on custom Apple Foundation Models built with Google's Gemini technology. Apple has published almost no detail on how the assistant selects or cites web sources, leaving brands with a new, currently unmeasurable answer surface to track.
- Google's AI Overviews criticize brands 44% more than ChatGPT, new data shows
A March 2026 BrightEdge analysis found Google's AI Overviews are 44% more likely to say something negative about a brand than ChatGPT. The more useful finding is where that negativity lands: ChatGPT surfaces it far closer to the purchase decision, even though it happens less often overall.
- Content length barely predicts AI citations. Here's what does
Ahrefs found almost no relationship between word count and getting cited in AI Overviews, a correlation of just 0.04 across 174,048 pages. The metric that actually predicts visibility, Position-Adjusted Word Count, measures something else entirely. Here's what the research says content teams should optimize for instead.
- What XML sitemaps actually do (and don't do) for AI search citations
OpenAI, Anthropic, and Perplexity's crawler documentation never mentions sitemaps as a discovery or citation mechanism. Bing is the outlier, tying sitemap freshness signals to what shows up in Copilot's AI-generated answers, though even Bing won't promise a citation. Here's what each vendor's own documentation actually says, and what a sitemap can and can't do for your GEO strategy.
- Agentic AI browsers like ChatGPT Atlas and Comet are changing what brand visibility means
ChatGPT Atlas, Perplexity Comet, and Microsoft's Copilot Mode don't just cite your content, they open the real page in a live, often signed-in browser session and act on it. That shift raises the stakes for brand accuracy and exposes security gaps several major vendors haven't fully closed. Here's what changes when an AI stops reading about your brand and starts browsing it directly.
- Amazon retired Rufus and launched Alexa for Shopping: what changed for brands
Amazon replaced Rufus with Alexa for Shopping in May 2026, merging its AI shopping assistant with Alexa+ into one system already reaching more than 300 million customers. The rename didn't change what decides whether a brand gets named out loud: a proprietary knowledge graph, a small set of large language models, and the product data feeding both. Here's what actually shifted, and how to show up in the answer.
- How to analyze AI crawler traffic in your server logs
Google Analytics never sees GPTBot, ClaudeBot, or PerplexityBot, because none of them load JavaScript. This is a practical guide to reading raw server logs instead: which crawlers actually matter, how to verify they aren't spoofed, and what Cloudflare's new September 2026 default-block policy changes for sites that depend on AI citations.
- Content cannibalization now costs you AI citations, not just Google rankings
Bing confirmed in December 2025 that near-duplicate pages on the same domain get grouped by AI models and reduced to one citation, sometimes the wrong one. New Ahrefs data shows most sites are already losing to this without realizing it. Here's how content cannibalization works once an AI engine is doing the ranking, and how to consolidate before it picks for you.
- A 126-million-prompt study exposes the gap between AI mentions and AI citations
Semrush's 2026 AI Visibility Index analyzed 126 million AI search prompts and found that being mentioned by an AI engine and being cited as its source are two different outcomes, with the gap ranging from 30% overlap on Gemini to 64% on Google AI Overviews. Only 36 brands out of more than 1,200 tracked held top-100 visibility across every major AI platform for four straight months.
- How to get cited by Grok, the AI engine that treats X posts as sources
Grok pulls nearly half its answer citations from X posts rather than websites, and its US chatbot share has grown into the double digits. Here's what xAI's own documentation, SpaceX's IPO filing, and independent citation research say about how to actually get cited.
- Why AI engines trust your Wikipedia page more than your own website
Wikipedia accounts for nearly half of ChatGPT's top-cited sources, and AI engines increasingly check brand facts against public records like Wikidata and Google's Knowledge Graph rather than a company's own website. This piece breaks down why entity data has become a quiet but critical layer of AI visibility, and the concrete steps to get yours right.
- The crawl-to-refer ratio: what it is, and why AI bots take far more than they give back
Cloudflare's own year-end data shows Anthropic's crawlers hit some sites roughly 500,000 times for every reader they sent back in 2025, and the imbalance grew all year. Here's what the crawl-to-refer ratio measures, why it's rising, and what it means for how you track AI visibility.
- Why your brand can look current in one AI engine and a year out of date in another
Anthropic's newest Claude model and its cheapest one are trained fourteen months apart, and the gap between AI vendors is wider still. This piece breaks down why AI engines disagree about how current your brand is, what live search does and doesn't fix, and what actually shortens the lag.
- Why Microsoft Copilot's 1.3% market share doesn't mean you can ignore it
Microsoft Copilot has the smallest share of any major AI chatbot, just 1.3% according to First Page Sage's September 2026 tracking, yet it runs inside Windows and Microsoft 365 for roughly 218 million active users. Because Copilot draws its citations from the same Bing index that powers ordinary search results, the GEO work required to get cited there overlaps almost entirely with standard Bing SEO. This piece breaks down how Copilot's retrieval actually works, what Microsoft's new Web IQ grounding API changes, and why blocking Bingbot removes a site from Copilot and Bing search at once.
- How AI engines actually chunk your content before they cite it
OpenAI's retrieval tool defaults to 800-token chunks; Microsoft's search recommends starting at 512. Then Google told publishers in May 2026 that chunking content is unnecessary for its own AI systems. Here's what actually determines whether a section gets cited, and which engines the advice does and doesn't apply to.
- Agentic commerce is growing fast, and most product pages aren't built for the AI agents buying from them
AI-referred traffic to U.S. retail sites grew 693% during the 2025 holiday season, and OpenAI's Instant Checkout now lets ChatGPT users buy without leaving the chat. Adobe's own data shows product pages score the lowest AI-readability of any page type on a retail site, which means the pages built to convert are often the hardest for shopping agents to parse.
- Do AI crawlers render JavaScript? What GPTBot, ClaudeBot, and PerplexityBot actually see
GPTBot, ClaudeBot, and PerplexityBot all read raw HTML and ignore anything that only renders after JavaScript runs, according to crawl data from Vercel and a real-world test from SEO analyst Glenn Gabe. Google's Gemini crawler is the exception, because it shares Googlebot's own rendering pipeline. Here's what each bot actually sees, and how to check whether yours is one of the pages left out.
- How often AI engines get brand facts wrong, and who's liable when they do
AI search engines get citations wrong more than 60% of the time, according to Columbia's Tow Center for Digital Journalism, and a Canadian tribunal has already ruled a company liable for what its chatbot invented. Here's what the data shows about brand-specific AI hallucinations, and how to catch one before a customer does.
- What robots.txt actually blocks when AI crawlers visit your site
OpenAI runs four crawlers, Anthropic runs three, and Perplexity runs two, each with a different job and its own robots.txt rule. This is the bot-by-bot reference for what disallowing GPTBot, ClaudeBot, or PerplexityBot actually stops, and what it leaves wide open.
- How REsimpli went from invisible to #1 in ChatGPT in 90 days
REsimpli, a CRM built for real estate investors, had zero presence in AI-generated answers for their category. Within 90 days they became the number one cited tool in ChatGPT for their core queries. This is what they changed, and why it worked.
- How to get cited by Claude AI: the complete 2026 playbook
Claude does not pull citations from Google. It retrieves live sources from Brave Search, a separate 40-billion-page index with its own ranking signals. This guide explains the two-layer strategy brands need to earn consistent citations from Anthropic's AI in 2026.
- GEO for B2B SaaS: how to build AI authority when you're a category-creating startup
51% of B2B software buyers now start their vendor research in an AI chatbot, not Google. For B2B SaaS companies, especially those creating a new category, that shift means your next customer may never see your website until after they've already made up their mind. This guide explains how to build AI authority from scratch and earn citations before the buyer ever hears your name.
- Answer-first content: the writing framework that gets AI citations
Most GEO guides tell you what to optimize. This one tells you exactly how to write, sentence by sentence, so your content becomes the answer AI engines extract. The framework operates at the paragraph level, not the page level.
- llms.txt: what it is, whether it works, and how to build one
Ahrefs analyzed 137,000 sites and found 97% of llms.txt files received zero traffic from AI bots in May 2026. Yet adoption grew 8.8x in twelve months. This guide explains what the file actually does, where it genuinely helps, and how to write one in under 30 minutes.
- LLM Brand Visibility vs Domain Authority: What AI Answers Reward
Domain authority predicts rankings in a list of links, not how often AI engines name a brand. Here is what the 2026 data shows about the gap, where authority still matters, and how to measure LLM brand visibility directly.
- How to Build the Business Case for AI Share of Voice
Marketing budgets face more scrutiny in 2026, and AI Share of Voice needs its own business case. This post lays out the three data points and the report structure that make leadership listen.