Field notes from the work.
Every post here traces back to a real build, a real number, or a study I went and checked myself, not a vendor's talking points. For the broader discipline of organic growth, see utog.blog ↗.
CDN and WAF rules that accidentally block AI bots
Cloudflare's AI-bot blocking is the most documented case because it ships as a visible, named toggle. Fastly, AWS WAF, and Akamai each offer AI-bot detection and blocking as customer-configured tooling rather than a named default, which makes the failure mode on those platforms less visible, not less possible: a managed bot-fighting rule, a generic rate limit, or a fingerprinting rule built for a different purpose can block an AI crawler just as silently.
Read the framework ↗Strategy & FrameworkDo AI browsers (Comet, Copilot, Atlas) change the citation game?
ChatGPT Atlas launched on macOS in October 2025; Perplexity's Comet went from a $200/month exclusive to completely free by March 2026; Microsoft and Google added AI browsing modes to Edge and Chrome the same window. Reported usage figures for all three vary wildly by source and aren't independently reconciled, but the category is real and growing, and it changes citation mechanics by making the browser itself the retrieval and citation layer rather than a separate chat interface.
Read the framework ↗Access & CrawlabilityDo AI crawlers even respect robots.txt?
Not consistently. A July 2026 study of 10,894 domains found 39.5% of robots.txt Disallow rules aimed at GPTBot specifically are not honored. The same study found sites that blocked at least one AI crawler got cited in Google AI Mode roughly 3x more often than the sample's baseline blocking rate, a finding that complicates the simple assumption that blocking keeps you both safe and invisible.
Read the framework ↗Access & CrawlabilityDo you need a separate sitemap for AI engines?
No. Your existing sitemap.xml already does the discovery job for AI crawlers that use it, and no major AI lab has published a requirement for a separate, AI-specific sitemap format. The genuinely new file in this space, llms.txt, is explicitly designed to complement a sitemap and robots.txt, not replace either, and Google's own John Mueller has stated no AI crawler has officially confirmed extracting information through it.
Read the framework ↗Strategy & FrameworkDoes Grok cite sources differently than the rest?
Grok cited sources in 27.01% of tracked responses in Superlines' 62-brand, 10-platform study, the highest rate among the platforms Superlines measured and far above ChatGPT's 0.59% in the same dataset. A separate, independent study found Grok 3 the least accurate of eight AI search engines tested for citation accuracy, wrong 94% of the time. Grok cites often and cites badly, by the two best-available independent datasets.
Read the framework ↗Strategy & FrameworkHow ChatGPT actually chooses which sources to cite
OpenAI hasn't published a scoring formula for how ChatGPT picks which fetched pages get an actual citation. What's confirmed: ChatGPT only searches the live web in a subset of conversations, it leans on Bing plus licensed media partners for retrieval, and multiple SEO-vendor analyses (unverified, but consistent) describe citation favoring extractable, directly-answering passages over top organic rank.
Read the framework ↗Strategy & FrameworkHow Claude's citation behavior differs from other models
Muck Rack's Generative Pulse data already shows Claude citing in only 55% of responses against ChatGPT's 96%, pulling from a noticeably different, more niche set of outlets. Anthropic's own documentation confirms the basic mechanism (Claude decides when to search, then cites the specific pages its answer actually references) without disclosing a selection algorithm, and vendor reports on how often Claude even invokes search at all vary widely.
Read the framework ↗Strategy & FrameworkHow Gemini surfaces brands in shopping and local queries
Google reported AI Mode passed one billion monthly users at I/O 2026, with shopping queries built on a Shopping Graph Google describes as roughly 50 billion products, around 2 billion updated hourly. Visibility depends on structured, fresh product feeds in Google Merchant Center and local stock data, not content optimization in the usual sense; a vendor-reported source mix (41% retailer, 37% earned media) is unverified but directionally consistent with how the broader AI-citation data on this site already breaks down.
Read the framework ↗Strategy & FrameworkHow Google AI Mode differs from AI Overviews
AI Overviews is a short generated summary sitting above Google's normal results page. AI Mode is a separate, full-page, multi-turn conversational experience that fans a query out into up to 16 parallel sub-queries. The two surfaces run on different Gemini model versions and, per Ahrefs' large-scale study, cite entirely different sources for the identical query roughly 87% of the time.
Read the framework ↗Strategy & FrameworkHow Google AI Overviews select and rank sources
Google's own documentation says AI Overviews need no optimization beyond standard SEO, and are grounded in Google Search's existing indexing and ranking rather than a separate algorithm. Beyond that, Google hasn't published a scoring formula. Vendor data on how closely citations track organic rank is directly contradictory, which is itself useful information about how little is actually confirmed.
Read the framework ↗Strategy & FrameworkHow Microsoft Copilot sources content (and why it leans on Bing)
Copilot grounds its answers in Bing's web index, crawled by BingBot. No separate, confirmed Copilot-specific crawler exists, which means Bing indexing health is the practical lever for Copilot visibility, the same dependency ChatGPT shares since it also draws on Bing. Microsoft Clarity data already on this site shows Copilot referrals converting at 17x the direct-traffic rate on subscriptions, the highest of any platform measured.
Read the framework ↗Strategy & FrameworkHow Perplexity sources and cites differently from ChatGPT
Perplexity is a search-first product built to cite by default; ChatGPT is a chat-first product where citation is a secondary mode. That design difference shows up in the data: a Columbia Journalism Review study found Perplexity misidentified a quote's source 37% of the time, the best score of any platform tested, and source-overlap estimates between the two platforms range from 11% to 44% depending on the dataset.
Read the framework ↗Access & CrawlabilityJavaScript rendering and AI crawlers: what gets missed
Vercel's analysis of over 500 million GPTBot fetches found zero evidence of JavaScript execution: GPTBot requests .js files as plain text roughly 11.5% of the time but never runs them. Independent checks from MERJ, Passionfruit, and Oncrawl confirm the same for ClaudeBot and PerplexityBot. Google's own crawlers are the one documented exception. For a client-rendered SPA, that's not a minor gap; it can mean the entire visible page is invisible to every major AI crawler except Google's.
Read the framework ↗Access & CrawlabilityPage speed and AI crawl budget: does it matter?
One secondary report claims OpenAI's crawlers abandon slow-responding pages with HTTP 499 errors on roughly 5% of requests; a server-log analysis from JetOctopus proposes a 600ms time-to-first-byte benchmark. Neither figure is independently verified, and the widely-repeated '2-3x more crawling for fast sites' multiplier traces to no disclosed dataset found in researching this piece. What is well-established, from Google's own documentation, is the general principle: a faster site lets any crawler fetch more pages in the same budget.
Read the framework ↗Access & CrawlabilityRobots.txt for AI crawlers: what to actually allow
A robots.txt that only handles the wildcard user-agent misses the point: AI labs run multiple distinct crawlers per company (training, search-index, and on-demand fetch bots), each with its own name, and a blanket rule treats them identically when you might not want to. This site's own robots.txt, allowing by name and using the emerging Content-Signal directive, is a working reference for what a deliberate AI-crawler policy actually looks like.
Read the framework ↗Content ExtractabilityStructured data for AI search: what actually gets used
This site's own matched study already found schema markup's pooled effect on AI citations statistically indistinguishable from zero. Breaking that pooled result down by individual schema type tells a more useful story: FAQPage's measured lift is small and contested, generic Article/Organization markup shows none, and the one type with a real, statistically significant effect is Product/Review schema carrying actual pricing and rating data.
Read the framework ↗Measurement & ROIWhat is 'share of model,' and why vendors are pushing the term
'Share of model' is a vendor-coined metric, not an industry standard: the percentage of AI answers in a category that name your brand, measured by running a fixed prompt set across models. It's a reasonable idea borrowed from share of voice, but the benchmarks being quoted around it vary by vendor, and the number itself is only as stable as the prompt set and models you ran it on.
Read the framework ↗Entity ResolutionWhat is a knowledge graph, and why AI engines lean on it
A knowledge graph is a database of entities (people, places, organizations, things) connected by labeled relationships, rather than a list of indexed pages. Google's own has grown from roughly 500 million objects and 3.5 billion facts at its 2012 launch to over 500 billion facts about 5 billion entities as of its last disclosed figure. The mechanics matter for AI search because an entity, not a page, is the unit an AI model is actually trying to identify before it can cite anything about it.
Read the framework ↗Entity ResolutionWhat is entity SEO, and why it matters more for AI search
Entity SEO is the pre-AI discipline of getting search engines to recognize a brand, person, or product as a single, well-defined entity, rather than just ranking for keywords. It predates AI search by over a decade, going back to Google's 2012 'things, not strings' Knowledge Graph launch. AI search didn't invent the discipline; it raised the stakes on getting it right, because a model has to resolve an entity before it can cite anything about it.
Read the framework ↗Measurement & ROIWhich AI platforms actually drive measurable traffic today
SE Ranking's referral-traffic data puts ChatGPT at 74.78% of all AI referral traffic, down from 79.74% a year earlier but still growing in raw volume. Similarweb's separate total-visit-share measure shows a much sharper ChatGPT decline, from 76.4% to 52.7% in twelve months, with Gemini's share more than tripling. The two datasets measure different things and aren't directly comparable, which is itself the most useful finding here.
Read the framework ↗Strategy & FrameworkWhat is an AI Shortlist Scan?
A plain, mechanical definition of the term: what actually happens across the 15 prompts, 4 engines, and 3 runs each, why that specific shape was chosen, and where it sits in a larger measurement program versus where it stops.
Read the framework ↗Measurement & ROIBuilding a fixed prompt set that reflects real buyer language
Most AI visibility prompt lists are keyword variations wearing a question mark. A prompt set that actually predicts anything starts from buyer intent (informational, comparison, recommendation, transactional, the same four categories IAB's own measurement framework formalizes) and gets filtered before it gets run.
Read the framework ↗Measurement & ROIDashboards that lie: how to spot a fake AI visibility metric
A precise-looking number isn't the same thing as a verified one. A real study found domain authority barely correlates with AI citations, and what circulates instead is a fabricated-sounding 0.81 correlation for a term no lab measures at all. Here's the five-question checklist that catches this pattern before it reaches a board deck.
Read the framework ↗Measurement & ROIHow many prompts do you actually need for a trustworthy read?
The University of St. Gallen's convergence analysis and SparkToro's ~3,000-run study both answer this from different angles, and they agree: a single prompt, run once, is a coin flip. Here's the actual arithmetic (prompts × runs × platforms × weeks) that separates a trustworthy read from a screenshot.
Read the framework ↗Measurement & ROIHow to catch factual drift in what AI models say about your brand
By the time a customer mentions a wrong fact an AI model told them about your pricing, it's already been repeated an unknown number of times. Here's how to build the monitoring habit that catches drift on your own schedule, using the same fixed prompt set the KPI framework already calls for, before a customer catches it for you.
Read the framework ↗Measurement & ROISetting up GA4 to catch AI referral traffic properly
GA4 added a native AI Assistant channel in 2026, and it still misses Perplexity, every session that arrives without a referrer, and most of ChatGPT's own app traffic. Here's the actual setup: what the native channel catches on its own, the custom channel group that catches more, and the floor you'll never close no matter how it's configured.
Read the framework ↗Strategy & FrameworkAEO vs GEO vs SEO: what's actually different
A mechanical, side-by-side comparison of what SEO, AEO, and GEO each actually optimize for: different retrieval systems, different success metrics, different surfaces, plus an honest look at where vendors blur the three together and why that isn't always wrong.
Read the framework ↗Strategy & FrameworkCan you pay to be cited by ChatGPT?
A clear, sourced yes/no: OpenAI's own self-serve ads launch versus Perplexity's full public reversal away from advertising, months apart, and what each company's own words say about whether money can buy a citation inside a generated answer.
Read the framework ↗Strategy & FrameworkDoes anyone actually own the term 'Answer Engine Optimization'?
A real attempt to trace who coined "Answer Engine Optimization" and when, and an honest report that it can't be pinned to one person, one company, or one date. Contrasted against GEO, which has a clean, verifiable origin story by comparison.
Read the framework ↗Strategy & FrameworkIs AI visibility a real discipline or a rebrand of SEO?
A genuinely evidence-led answer, not a defense of the industry that sells this work. What's mechanically new about AI visibility (crawler access, answer variance, citation-without-click) versus what's just entity SEO and off-site authority wearing a new acronym.
Read the framework ↗Strategy & FrameworkIs your brand on the AI buyer's shortlist before anyone even searches?
A specific, disclosed-methodology finding that a brand named inside ChatGPT's own generated search query reaches the final answer far more often than one only fetched during the search: what it means, its real limitations, and why most "get on the AI shortlist" advice skips the actual mechanism.
Read the framework ↗Strategy & FrameworkWhat is a generative engine, technically?
A plain technical definition of "generative engine" (the retrieval-and-synthesis pipeline behind ChatGPT, Perplexity, and Google's AI Overviews), and exactly how it differs mechanically from the index-and-rank pipeline behind a traditional search engine.
Read the framework ↗Strategy & FrameworkWhat is AEO (Answer Engine Optimization)?
A working definition of Answer Engine Optimization, without the vendor gloss: what it actually means to optimize for an answer instead of a ranking, where the term overlaps with GEO and SEO, and why no two vendors define it quite the same way. Anchored to what's verifiable, not what's sold.
Read the framework ↗Strategy & FrameworkWhat is GEO (Generative Engine Optimization)?
A working definition of Generative Engine Optimization, distinguishing the narrow academic term (a specific 2023 Princeton/Georgia Tech paper) from the much broader marketing umbrella the industry now uses it as. Both are real, they're just not the same size of claim.
Read the framework ↗Strategy & FrameworkWill AI search ever fully replace traditional search?
A big-picture, honestly-labeled answer: Google's actual current search share, ChatGPT's real growth trajectory, and the structural reasons full replacement is a weaker bet than fragmentation. Closes the AEO/GEO/SEO definitional cluster on this site.
Read the framework ↗Strategy & FrameworkBeing cited by AI and being chosen by it are different problems
A citation is a single event. Being the brand an AI system keeps recommending is a different, more durable state, and a 14,000-call study shows how durable it actually is depends entirely on what the person was asking, not on how well-optimized your content is.
Read the framework ↗Content ExtractabilityDoes schema markup help AI citations?
A controlled, matched study of 1,885 pages found schema markup's effect on AI citations statistically indistinguishable from zero overall, but a separate study that isolated schema types found a real, statistically significant exception for schema carrying actual data. The lift was never in the wrapper; it was in whether the wrapper had anything worth citing inside it.
Read the framework ↗Strategy & FrameworkHow long a wrong fact about your brand survives in AI answers
A citation can go stale on a schedule: the next retrieval either finds the correction or it doesn't. A mention sourced from training data has no schedule at all. Real knowledge-cutoff dates show how long that gap can actually run, and why fixing the source today doesn't retroactively fix what a model already memorized.
Read the framework ↗Measurement & ROITracking your AI consideration set, not just your citation count
Whether you're even in the 3-5 brands a model compares matters more than whether you win the comparison. Here's how to actually track it, and why a 14,000-call study means it has to be tracked by query intent, not as one blended number.
Read the framework ↗Entity ResolutionWhat to do when an AI model gets a fact about your brand wrong
The thumbs-down button on ChatGPT, Gemini, and Perplexity doesn't correct the answer you just saw. On all three platforms, feedback feeds an aggregate training pipeline, not an in-place edit. Here's what actually fixes a wrong AI answer, and why it depends on whether the error came from retrieval or training data.
Read the framework ↗Measurement & ROIIAB's AI visibility measurement framework, explained
In August 2026, IAB published a formal measurement framework for AI visibility: the industry's first attempt at a shared vocabulary, a two-tier data-quality standard, and a required-disclosure checklist for the 20+ companies now selling this data. Here's what it actually standardizes, where it lines up with definitions already used on this site, and where it draws the line slightly differently.
Read the framework ↗Entity ResolutionWhy LLMs can't pick a random number, and what it means for AI visibility
A Harvard-authored ACL 2026 paper found 10 of 11 frontier models fail every statistical test for random sampling when asked one prompt at a time. Four independent studies since 2024 keep landing on the same handful of numbers. Here's why it happens, and what the same mechanism means for whether AI recommends your brand at all.
Read the framework ↗Entity ResolutionAI visibility for banks: why regulated brands are losing the answer layer
Nearly half of global consumers now use AI to guide savings and investment decisions, and the Big Four US banks hold 77% of deposits but under 15% of AI citation share. What AI visibility actually means for a bank or NBFC, why the gap exists structurally, and why India's regulator is already treating it as a governance question, not just a marketing one.
Read the framework ↗Measurement & ROIDo AI visibility audits actually predict anything?
Not 'is the number accurate': a narrower, more useful question. Does a high or low AI visibility score today predict anything real tomorrow, more citations, more leads, more revenue? No study meeting basic disclosure standards was found that tests this directly. Here's why the honest answer is 'unconfirmed,' not 'no' or 'yes.'
Read the framework ↗Off-site & EarnedHow fast review velocity actually moves AI visibility
Two real G2 studies look like they disagree. One finds a 4-day citation lift from a review burst, the other finds total review count barely predicts citations at all. They don't actually contradict each other. Here's the mechanism that reconciles them: velocity and recency, not lifetime volume, are what G2's own citation-relevant scoring rewards.
Read the framework ↗Off-site & EarnedHow to actually earn a citation from a journalist
Journalism alone accounts for 25-27% of all AI citations, consistently, across three Muck Rack studies. Getting that coverage isn't a pitching problem first: it's a "do you have anything worth citing" problem. Here's what actually earns a journalist's attention, backed by disclosed survey data, before the pitch ever gets sent.
Read the framework ↗Off-site & EarnedHow to pitch a journalist for a citation-worthy mention
The tactical execution guide: where journalist requests actually live now that HARO has changed hands twice, what response rates really look like, and the exact format (length, subject line, single ask) the disclosed data says gets a reply. Assumes you already have something worth pitching.
Read the framework ↗Content ExtractabilityHow to write a single extractable answer block for a cornerstone page
A tactical micro-guide, not a philosophy: the exact recipe for the one block every cornerstone page needs, where it goes, how long it runs, and a checklist to audit whether yours actually qualifies. Uses real, published answer blocks from this site as worked examples instead of invented ones.
Read the framework ↗Strategy & FrameworkMention vs. citation vs. recommendation in AI search
"AI visibility" gets treated as one metric when it's actually three separate outcomes that don't move together: a citation, a mention, and a recommendation. Here's the precise difference between them, why they're measured differently, and which one actually matters for a given goal.
Read the framework ↗Measurement & ROIThe AI visibility score industry: who's really being served
$154.5 million into one AI-visibility platform in roughly 18 months, chasing a metric that a rigorous academic study found swings 65% day to day. That combination isn't necessarily fraud, but it's worth separating the verified facts from the interpretation, and being honest about which is which.
Read the framework ↗Strategy & FrameworkThe most repeated fake statistic in AI search marketing
A real study found domain authority has essentially no relationship to AI citations: a plain, honest null result. What actually circulates instead is a cluster of oddly precise numbers (a 0.18-versus-0.81 correlation, specific "trust score" thresholds) that don't trace back to that study, or to any other named, disclosed source. Here's the real finding, and the fabricated one that replaced it.
Read the framework ↗Access & CrawlabilityWhat is llms.txt? The actual file format, explained
A reference, not an argument: what an llms.txt file actually contains, the exact Markdown structure the spec defines, how it differs from robots.txt and sitemap.xml, and where llms-full.txt fits in. For whether any of this moves AI citations, see the dedicated deep dive.
Read the framework ↗Measurement & ROIWhat vendors won't tell you about their own AI-visibility methodology
AI visibility score already asks what to check before trusting a score. This is the fuller version: a direct comparison of what three real, funded vendors actually disclose about their own methodology, from the most transparent to the least, and why that gap matters more than which one has the biggest dataset.
Read the framework ↗Strategy & FrameworkAI Overviews vs. featured snippets: what's actually different
They look similar at the top of a Google results page, but they're mechanically different systems with different source counts, different trigger rates, and, per real click-tracking data, a meaningfully bigger hit to click-through. Here's the actual comparison, including a widely-repeated claim about the two that nobody can source.
Read the framework ↗Access & CrawlabilityDoes llms.txt actually get read by anyone?
A sharper, standalone answer to a question this site has already flagged twice: 97% of published llms.txt files get zero AI-crawler requests in a month, Google has now said in writing that Search ignores the file entirely, and even the small counter-evidence that exists doesn't move the conclusion. Here's the full picture, including where the file genuinely does help.
Read the framework ↗Measurement & ROIHow to build your own AI visibility tracking without a vendor tool
A practical walkthrough of what it actually takes to build your own AI visibility tracker: the real API options per platform, real costs, and the one platform you genuinely cannot replicate this way. Extends the fixed-prompt-set methodology already covered in the AI visibility KPIs post into an actual build.
Read the framework ↗Strategy & FrameworkWhat a 90-day AI visibility roadmap actually looks like
A concrete, week-by-week roadmap built directly on the five-stage sequence already published here, with an honest caveat up front: 90 days gets the foundation solid and the compounding lever started, not full off-site dominance. Anyone promising the latter on a fixed timeline is selling past what any disclosed study supports.
Read the framework ↗Strategy & FrameworkWhat does it mean to be "cited" by an AI model?
"Citation" isn't one consistent thing across AI products, and being read isn't the same as being credited. A technical breakdown of what actually happens between a model fetching a page and a citation appearing, including real data on how often a source gets used without ever being named.
Read the framework ↗Strategy & FrameworkWhere "X% of people use AI search" stats actually come from
A source-tracing piece: five widely-repeated "X% of people use AI search" statistics, each traced back to what its primary source actually measured. Most get generalized well past what the underlying study says. One holds up exactly as quoted.
Read the framework ↗Off-site & EarnedWhy earned media outperforms owned content in AI citations
Across three independent studies with disclosed methodology, earned media (reviews, journalism, third-party coverage) accounts for 82-89% of AI citations. Owned content, including your own blog, is the minority channel. This piece lays out the full body of evidence, plus what's actually known about why.
Read the framework ↗Access & CrawlabilityCloudflare's AI bot blocking: what it silently breaks
A permissive robots.txt doesn't mean AI crawlers can reach you. Cloudflare's managed AI bot block rules sit in front of your application and can override it silently: here's exactly what breaks, why it's invisible from your own dashboard sometimes, and how to check.
Read the framework ↗Entity ResolutionG2/Capterra profile hygiene for AI visibility
G2's own data shows 80% of its products get more AI citations than human pageviews, but only if the profile is actually optimized and complete. Category placement, field completeness, and now a real consolidation event (G2 acquiring Capterra) all change what profile optimization means in 2026.
Read the framework ↗Content ExtractabilityHow to write content that AI models can actually quote
How to write for AI search, backed by the actual GEO research. Adding statistics, citations, and quotations is the strongest documented lever, but "add more stats" isn't a writing technique on its own. Here's what extractable writing actually looks like, with real before/after rewrites.
Read the framework ↗Content ExtractabilityThe GEO paper (Princeton/Georgia Tech): what it actually proved
The GEO paper, explained properly, cited three times elsewhere on this site as shorthand for "add statistics and citations." Here's the actual paper: the nine methods it tested, the two metrics it measured, the domain-dependent finding that gets dropped in most summaries, and why keyword stuffing was one of the methods it tested and found didn't work.
Read the framework ↗Entity ResolutionWhat is entity consistency, and why AI engines need it
A model doesn't cite a name, it cites an entity it has confidently resolved. Here's what "entity consistency" actually means mechanically, why disambiguation is the real bottleneck, and what breaks when your brand resolves as three different nodes instead of one.
Read the framework ↗Entity ResolutionWikidata for brands: is it worth setting up?
Yes, if you qualify: Wikidata for brands is the single highest-impact entity signal because it feeds directly into Google's Knowledge Graph rather than being one source among many. This piece covers the actual qualification bar, the real effort involved, and what to do if you don't clear it yet.
Read the framework ↗Entity ResolutionEntity audit in under an hour
A quick, no-tool-required entity audit checklist for whether your brand resolves as one clean entity to the systems AI models cross-reference, before you spend on anything else in Stage 2.
Read the framework ↗Access & CrawlabilityGPTBot, ClaudeBot, PerplexityBot: how to audit your access logs
A user-agent string is a claim, not proof. Here's the actual reference table of major AI crawler user agents, why matching on the string alone isn't enough, and the verification method (FCrDNS) that catches a spoofed one.
Read the framework ↗Access & CrawlabilityHow to check AI crawler access
Checking robots.txt tells you what you're asking AI crawlers to do, not what's actually happening. How to check AI crawler access with a three-layer verification (robots.txt, server logs, and the CDN/WAF layer that can silently override both), plus real data on how unreliable "blocked" actually is in practice.
Read the framework ↗Access & CrawlabilityServer log analysis for AI visibility: a practical guide
Your analytics tool doesn't see AI bots. They don't execute JavaScript, so nothing fires. Raw server logs are the only place this traffic is visible at all. Here's the actual method: where the files live, what to extract, and how to turn a one-off log analysis into a repeatable report.
Read the framework ↗Measurement & ROIAI visibility score
An AI visibility score is one draw from a wide, wobbly distribution, not a fixed number. Here's how these scores are actually calculated, what a 45-day academic study found about how much they swing day to day and platform to platform, and how to tell a trustworthy one from a guess with a decimal point.
Read the framework ↗Measurement & ROIAI visibility KPIs
Most "AI visibility KPI" lists are eight-metric wish lists with no measurement discipline behind them. Here's what's actually stable enough to track, what's structurally noise, and the sampling rigor that separates the two.
Read the framework ↗Measurement & ROIAI visibility ROI
AI visibility ROI gets dismissed as unmeasurable or oversold with cherry-picked multipliers, often in the same week. Here's why most published "X times higher conversion" numbers aren't comparable, what the disclosed-methodology studies actually show, and how to fix the attribution gap that undercounts AI-referred traffic in the first place.
Read the framework ↗Strategy & FrameworkAI visibility strategy
An AI visibility strategy is a sequenced system for getting cited by AI answer engines, not a monitoring dashboard. What it actually is, why most 2026 playbooks get the order backwards, and how to build one that holds up under primary research.
Read the framework ↗Strategy & FrameworkThe five stages for winning in the AI search era
A one-time AI visibility score isn't a strategy. Here are the five stages (in the order they actually have to happen) for winning in the AI search era, with the primary research behind each one.
Read the framework ↗Your competitor is the answer
every AI gives your buyers.
A $497 Shortlist Scan tells you for certain in three days: real buying-stage prompts, run cold, three times each across ChatGPT, Perplexity, Gemini, and Claude.