Research

14 articles

Research
ChatGPT ads are losing to ChatGPT’s free recommendations
The first large study of ChatGPT ads found a paid click earns about a third of a Google search click. For the same shops, ChatGPT’s free recommendations did far better.
Ali KhalladNov 26 min read
Research
ChatGPT no longer reads your page before recommending you
ChatGPT used our pages in 9 of 10 answers and fetched one page, once. Cloudflare data shows its page fetches down 70% from the June peak.
Ali KhalladSep 286 min read
Research
Fake AI crawlers are scanning for Claude and OpenAI API keys
We checked every visit that arrived with an AI crawler name against the address lists OpenAI, Anthropic, Google, Microsoft and Perplexity publish. More than a third were not who they said.
Ali KhalladSep 86 min read
Research
How many checks before an AI visibility score means anything?
Of the combinations where a brand appeared at least once, 80.4% were inconsistent. On those, a single check agrees with the majority verdict only 72.2% of the time.
Ali KhalladSep 74 min read
Research
ChatGPT citations dropped Reddit. Perplexity and Gemini didn’t.
We ran a controlled before-and-after on our own tracked set, same brands and same questions in both windows. ChatGPT’s Reddit citations collapsed. Perplexity and Gemini did not follow.
Ali KhalladSep 72 min read
Research
What replaced Reddit in ChatGPT’s answers
Reddit fell from 34.4 percent of ChatGPT answers to 3.2 percent in a single week in August 2026. Government and regulatory sources more than tripled to fill the space, and everything else shrank.
Ali KhalladAug 204 min read
Research
I told eight AI assistants not to read a page. Three read it anyway.
I blocked one page in robots.txt and asked eight AI assistants to read it. Three said no, three read it anyway, and two never showed up. The split is not what I expected.
Ali KhalladAug 197 min read
Research
Same question, different sources: how much AI citations change between checks
Ask an AI the same buyer question repeatedly and the cited sources shift. We measured how much they change between checks, and it turns on the engine and the question.
Ali KhalladJul 225 min read
Research
Brands are missing from most AI answers. Here is who gets cited.
We logged about 26,400 AI answers across five assistants. Brands were cited or named in only 16 percent. Here is the per-engine split and why.
Ali KhalladJul 166 min read
Research
Where ChatGPT, Perplexity, and Gemini Actually Get Their Information
We classified 151,000 sources five AI engines actually cited. The open web dominates every engine, but the leans differ: YouTube on Google AI Mode, forums on ChatGPT, almost neither on Claude.
Ali KhalladJul 84 min read
Research
When AI Describes Your Brand, It Cites Your Competitors
We analyzed almost 100,000 AI citations across five platforms. Of the outside sources AI cites about a brand, around 40 percent are its competitors.
Ali KhalladJun 296 min read
Research
We Analyzed 127,198 AI Citations. The Five Engines Barely Read the Same Web.
An experiment across about 16,400 AI answers and 127,198 source citations from ChatGPT, Claude, Gemini, Perplexity, and Google AI Mode. The five engines agree on just 2.7% of sources, and Gemini cites three times more than ChatGPT.
Ali KhalladJun 275 min read
Research
Google Search Is Becoming an AI Agent
Google Search is moving from links to answers, agents, and generated interfaces. That changes what web visibility means.
Ali KhalladMay 208 min read
Research
What nginx logs prove about AI traffic vs referral traffic
I ran an nginx probe against ChatGPT, Claude, Perplexity, Gemini, Copilot, Grok, Meta AI, and Manus to see what each assistant actually sends when it fetches your page. Real user-agents, real Accept headers, real vendor blind spots.
Ali KhalladApr 198 min read