September 2026 Summaries

AI people search tools help sales and recruiting teams identify professionals using criteria such as job role, company, location, expertise, and business contact information. The overview distinguishes among B2B professional-search platforms, conversational web-search tools, and platform-native services such as LinkedIn Sales Navigator, comparing Exa, AI Ark, Jenova, HyperWrite, and LinkedIn by inputs, outputs, and pricing. It describes a typical workflow of defining target criteria in natural language, generating and validating a list of candidates, and enriching records with selected fields such as titles, profile URLs, emails, or phone numbers. Exa promotes its Agent API as a tool for producing schema-validated, cited JSON results from an index it says contains more than one billion person records, with separate charges for contact enrichment. The piece also advises users conducting name-based research to provide identifying context and verify source profiles, since web-based summaries can be incomplete or may conflate people with the same name.

Grok, xAI’s AI assistant, supports real-time retrieval from the web and public X posts through built-in tools, allowing it to cite sources and provide current information beyond its May 2026 knowledge cutoff. Its DeepSearch agent, introduced with Grok 3 in February 2025, gathers and evaluates information to generate concise reports, while developers can access web_search and x_search through the xAI API. Web search API calls cost $5 per 1,000 requests plus tokens, and X Search has separate charges for fetched posts and profiles; users can also apply limited domain filters, though Grok may still produce confident inaccuracies. The Exa team presents its search API as a portable alternative that works across Grok, Claude, GPT, and Gemini, costs $7 per 1,000 requests, and can be connected to Grok through Remote MCP tools or the Grok Build marketplace.

Claude supports real-time web search across its applications, API, and Claude Code, allowing it to retrieve current information, fetch provided URLs, and cite sources in search-grounded responses. Anthropic’s system can autonomously determine when current information is needed or be directed through prompts, while users can enable the feature manually in some chat interfaces and organizations may need to activate it through administrative settings. In the API, developers add a web search tool and can control search limits and domain restrictions, with search priced at $10 per 1,000 queries plus token costs, while web fetching incurs token costs only. Claude Code includes WebSearch and WebFetch tools, although Bedrock users may need external integrations. The piece also presents Exa as an alternative search provider that can connect through MCP, supplying concise source highlights intended to conserve context-window capacity and offering full-page retrieval when needed.

AI source finders streamline academic research by searching scholarly databases and the web for material supporting a topic or draft, summarizing relevant sources, and generating formatted citations, though users must verify every result. The comparison covers Exa, Sourcely, JotBot, Consensus, Elicit, Perplexity, and Scite, which differ in their source coverage, citation capabilities, free allowances, pricing, and specialized features such as consensus measures, sentence-level citations, source chat, and citation-context analysis. Exa searches a dedicated index of roughly 350 million publications alongside the open web and offers APIs that return supporting passages and publication metadata, allowing developers to build source-finding tools for claims, longer drafts, and structured research tasks. The guidance emphasizes confirming that cited works exist, reading the exact supporting passage, checking whether findings have been challenged or retracted, and reviewing citation formatting against the required style guide.

Five major AI chatbots—ChatGPT, Perplexity, Microsoft 365 Copilot Chat, Gemini, and Claude—offer web search capabilities in 2026, though they differ in availability, automatic search behavior, query transparency, and citation formats. Web-enabled chatbots generally determine whether a question needs current information, convert it into one or more search queries, retrieve and filter relevant passages, and synthesize a cited answer from those sources. ChatGPT and Claude typically decide when to search but can be instructed to do so, while Perplexity searches every query by default. Exa positions its MCP server, APIs, and SDK tools as options for replacing or adding web search in compatible assistants and custom chatbots, emphasizing precision-focused retrieval that supplies only relevant, cited information rather than entire webpages.

Exa’s guide surveys 12 research-paper search engines in 2026, grouping them into broad multidisciplinary tools and specialized databases while recommending combinations by discipline. General services such as Google Scholar, Semantic Scholar, BASE, CORE, Dimensions, and Exa offer wide coverage, citation features, open-access discovery, or links between publications, grants, datasets, and policy documents, whereas PubMed, IEEE Xplore, JSTOR, arXiv, SSRN, and ERIC serve fields including medicine, engineering, humanities, preprints, social sciences, and education. The guide emphasizes that researchers benefit from using both broad and subject-specific sources to improve coverage and access full text. It also presents Exa’s API-based publication index of roughly 350 million records as an option for developers building semantic research tools, highlighting its natural-language search, cited metadata, and passages supporting returned results.

Exa has introduced Agent Ultra, its highest-effort research mode, which coordinates multiple agents across thousands of sources to conduct exhaustive deep research, entity enrichment, and list-building tasks. The company reports that Agent Ultra surpassed Opus 5.5, GPT-6 Astra, and Perplexity Agent on benchmarks including WANDR, DeepSearchQA, WideSearch, and Find-All Company, while claiming lower costs per task or entity found. Designed for applications such as assembling model-training data, financial due diligence, KYC research, market mapping, monitoring business signals, and go-to-market account development, the system breaks complex requests into subtasks and assigns appropriate models based on required depth. Agent Ultra is available through Exa’s API and Playground, alongside the company’s broader search, crawling, content, monitoring, and integration products.

Exa has introduced Snapshot, a research-preview capability that provides a versioned view of the web through more than 400 billion webpage snapshots collected over two decades. Available through the Search and Contents APIs via the `snapshotAsOf` parameter, it returns search results and webpage content as they existed on a specified date. The feature is intended to support reproducible AI evaluations by preventing agents from finding solutions that appeared online after a task was created, helping distinguish genuine problem solving from copied answers. Snapshot also supports financial and quantitative research by enabling point-in-time web-data backtesting without requiring extensive historical data collection. Users can access it through Exa’s dashboard or APIs, while enterprise details such as full index coverage, rate limits, and zero-data retention are available through Exa’s team.