Notes on SEO monitoring, AI SEO performance, and how to keep tabs on your brand.
Exa is the better search API for most AI agents: in our benchmark it answered more factual and multi-hop questions correctly than Tavily and returned results in 0.74 seconds against 2.00. Tavily costs less per correct answer on a 100,000-search plan.
We gave cloro, SerpApi, Tavily, Exa, Parallel and Brave Search API the same 2,182 questions, one search each, and graded every answer. Then we tested which parts of a search result an AI agent needs to answer correctly.
Five SERP APIs ship both an official Python SDK and an official MCP server. The package names, the hosted endpoints, how each one authenticates, and the live price per 1,000 Google searches.
We sent 96 prompts to Bing AI Mode and Google AI Mode. Only 86 of 891 domains appeared in both, and Google cited Reddit in 25 answers to Bing's 1.
How to scrape Bing Copilot Search with Python and Playwright: read the answer iframe, its sources and citation chips, tested on 12 live pages.
Copilot rank tracking checks if Bing Copilot names your brand and cites your pages. Set it up with Bing Webmaster Tools, manual checks or an API.
Two ways to give a language model live web search: search first and put the results in the prompt, or give the model a search tool. Python code for both, a TypeScript agent, and what we measured on accuracy, latency and cost.
We ran 10 prompts through Bing Copilot APIs and compared 8 Copilot scrapers on the page they read, speed, citations, targeting and price per 1,000.
Google goto URLs hide where a search result leads. Google began holding them on 30 October and changed them again on 6 November. How SERP APIs compare.