AI Search Optimization
insights
Research, guides, and practical playbooks for understanding how brands appear in ChatGPT, Gemini, Claude, and other AI answer engines.
Travel is where AI assistants meet the booking decision
Instinct says travel drives about half its agent transaction volume. Aiso's conversation examples reveal the decisions travelers ask AI to help make before booking.
Why AI mentions a brand: what 34,960 answers tell us
Across 34,960 GPT and Gemini answers, own-site citations and past visibility tracked brand mentions. Aiso’s sixth paper explains the findings and limits.
We ran the same prompts through four AI engines. They barely cited the same pages.
Our fifth research paper compares 589 citations from ChatGPT, Copilot, Google and Perplexity. Same-prompt URL overlap averaged 0.8%, and 85% of engine pairs shared no cited URL.
We gave ChatGPT a celiac memory. The searches stayed the same.
In a controlled NYC hotel test, dietary memory appeared in 60% of distinct personalized answers but 0% of observed fan-out queries.
Why does ChatGPT keep picking 17 when you ask for a random number?
We sampled 207,132 web pages to test whether 17 wins because it is rarer on the web. The data points to a learned idea of what random looks like instead.
How often does AI actually change a buying decision? We looked at real conversations
Our fourth arXiv paper audits real AI buying conversations. Recommendations appeared in 77.6% of purchase-directed episodes, while same-mission follow-up appeared in only 26.9%.
Chrome, Brave and Firefox: 9 AI-agent tasks, 9 completions
The same three tasks completed in actual Chrome, Brave and Firefox with zero action retries. Recorded timings, screenshots and limitations from a small Linux VM experiment.
We asked a ChatGPT agent to use Chrome, Brave and Firefox. Only one was available
An Aiso test found 100% task completion in Chrome with zero retries, but ChatGPT's cloud environment exposed only one of three requested browser families.
What does the new ChatGPT model mean for marketers?
GPT-6 Astra makes 68% fewer visual misses than GPT-5.6 Sol by Aiso's calculation. Why website design, visible trust signals, performance and agent-ready UX now matter for AI visibility.
ChatGPT ads start contextual, not personalized: what marketers need to know
OpenAI's August privacy update says the initial rollout uses the current conversation and limited context, not past chats or memories. Aiso separates the official disclosure from the marketer implications: conversation-level relevance, landing-page alignment, aggregate reporting, and the important unknowns.
The conversation is the real AI search query: evidence from three studies
Three Aiso papers connect the hidden search work inside AI answers, the intent that develops after the opening prompt, and the conversation context that materially changes the final response. A new measurement framework for conversational demand.
The prompt is not the query: how request state evolves across AI conversations
Our second arXiv paper measures how a user's request spreads across the turns of an AI conversation. Across 8,133 conversations the final prompt carries a median 36% of the session's user vocabulary, and about half the time a stated need never reaches the final prompt. Why AI-search demand has to be measured at the session, not the query.
How much search is hidden in one AI answer? Introducing search density
Our first arXiv paper measures the query and source work compressed into one conversational answer. The median information-seeking answer packs 11 retrievable facts into 3 facets, and a 6-case web pilot shows one broad query covers 70% of it. What answer-reconstruction search density means for measuring AI demand.
Luxury's AI gap: 82% of top buyers use AI, and it rarely cites the brand
Bain's 2026 luxury study found 82% of top spenders used AI in their last purchase, but 90% of the sources AI cites for luxury queries are third-party. Aiso's analysis of the demand-supply gap, with new ratios and conclusions for luxury marketers.
How GEO became the default name for AI search optimization
We asked four AI systems to name the practice of optimizing for AI answers. Three led with GEO. None led with AEO, even when citing a vendor who prefers it. Here's how one term won—and what it teaches about influencing the machines.
How LinkedIn's Feed Algorithm Gets Posts to Millions of Views
LinkedIn's March 2026 feed-engineering post says the feed ranks millions of posts, refreshes embeddings within minutes, and processes 1,000+ prior interactions. Here is what that means for writing posts that recommendation systems and LLMs can understand.
ChatGPT-User accounted for 47.4% of travel bot requests in a Cloudflare snapshot
How to read a dated Cloudflare travel benchmark without confusing user-action bot requests with people, recommendations or bookings.
Everything you trust is a black box
The most common objection to AI is that nobody can fully explain how it works. True, but name one thing you use every day that you can. On Bergson's intuition vs analysis, tacit knowledge, and how people already let AI pick their next purchase, plus real anonymized ChatGPT conversations where shoppers ask the model to just decide.
How many AI-search prompts to track: cover decisions, not wording variants
Build a prompt set around products, markets and buyer constraints, then budget repeats separately from coverage.
How often to run AI-search prompts: choose precision before cadence
Use repeated answers to estimate variation, without treating one, ten or forty daily runs as a universal rule.
The Brave of AI search: why full opt-in conversation data wins
Brave built an independent search index on strictly opt-in, anonymized, k-anonymous browsing data. Aiso applies the same principle to AI search: every conversation in our 5M+ IP panel is voluntarily shared, anonymized, and reported only in aggregate. The analogy, the mechanics, and why consent-first data is also better data.
Why one ChatGPT query tells you almost nothing about your brand visibility
We ran 19 real consumer prompts through ChatGPT ten times each under identical conditions. Two runs share only 55% of recommended brands on average. 27% are 'ghosts' - mentioned in one run and never again. Only 16% are stable across all ten runs. Why single-screenshot AI visibility audits are statistically indefensible.
Is your portfolio company winning AI search? 4 metrics that matter (and 1 that lies)
A scorecard for investors evaluating portfolio companies' AI search performance - and for the marketing leaders who'll get the question. Commercial-intent mentions, the referral underestimation problem, visibility per dollar, and revenue per dollar, with public benchmarks plus original data from Aiso's panel of millions of real anonymized AI prompts and a live gpt-5.3-chat-latest replay.
Bing just redefined the index: grounding is now the unit of value, not pages
Microsoft Bing's May 2026 post draws a sharp line between traditional search and grounding for AI answers. The five dimensions Bing now uses to measure index quality - and what brands should change.
LLM Ranking Factors: What Actually Determines Whether AI Recommends Your Brand
We analyzed millions of AI answers to uncover the top factors influencing brand recommendations across ChatGPT, Gemini, and Claude.
How much ChatGPT traffic should a brand expect?
Use a measured baseline and a relevant comparison period, not another company’s growth chart as your traffic forecast.
ChatGPT model activity in Q1 2026: what the usage distribution does and does not show
A dated model-activity snapshot with a derived comparison of the two largest categories, and why usage does not isolate deliberate user preference.
What Are the Real Sources Used by ChatGPT? A Detective Investigation
We isolated unique phrases and traced them back to Reddit, directories, and training data to reveal AI's real source mix.
Our ChatGPT Sample Demographics: Overview & Limitations
An honest look at the consent-based panel powering Aiso - who's in it, what biases to know, and how we mitigate them.
Testing ChatGPT's Non-Ranking Nature: Espresso Machines
We ran 400 queries to test whether ChatGPT maintains consistent rankings - and found something more nuanced.
ChatGPT’s search market share depends on what you count
Keep web visits, search-like questions and referral clicks separate before comparing ChatGPT with a search engine.
Comparing AI assistants: test the product, not just the model name
Compare ChatGPT, Gemini, Claude, Perplexity and DeepSeek using the same buyer tasks, search settings and scoring rules. Keep product usage separate from model capability.
What Are People Searching for on ChatGPT?
Real data on the most frequent prompts, products, and brands people query inside ChatGPT.
ChatGPT search privacy: distinguish a query from a conversation
Inspect what the search query contains before claiming a prompt leak. Separate documented provider sharing, captured query strings and Search Console evidence.
How AI Question Types Determine Citation Sources
Different prompt types pull from different sources. Here's the mapping for research, comparison, how-to, and purchase queries.
AI chatbot traffic conversion: compare orders, not just engagement
Use a sourced conversion benchmark and a clear break-even calculation without mistaking longer sessions for more sales.
How We Test and Run Experiments at Aiso
Our scientific commitments - falsificationism, hypothesis-driven design, replication - applied to AI search research.
AI referral traffic is growing, but the sector benchmark matters
A sourced comparison of U.S. travel and retail referral growth, with a calculation that keeps different starting bases separate.
Advertising in ChatGPT: The Next Frontier of Marketing
How OpenAI's monetization roadmap will reshape paid acquisition - and how to prepare your team and content.