Research by Ali Abouelatta
Field notes.
What agents can find. What they can use. What the evidence actually says.
Cofounder of Lazyweb (a16z Speedrun). Previously grew subscriptions and built native ads at Duolingo. Writes First1000 (90,000+ subscribers).
research-article · 2026-09-19
AEO vs SEO: what the studies actually measure
A source-led decision guide separating SEO eligibility, AI visibility, citations, clicks, and conversion using Google, Pew, Seer’s 2026 cohort, and the KDD 2024 GEO paper.
research-reading-list · 2026-09-19
Profound’s AEO research: a reading list with the disagreements left in
A skeptical, source-led reading list for Profound’s prompt, citation, volatility, intent, structural, and Reddit studies, with exact reported numbers and methodological limits.
product-review-research · 2026-09-19
Profound, Gauge, and AirOps: what the public reviews actually support
A bounded review-based comparison of Profound, Gauge, and AirOps using full TrustRadius evidence, clearly relevant Gauge AI reviews, Product Hunt feedback, and AirOps current-user excerpts without declaring a universal winner.
research-methods-article · 2026-09-19
If AI citations move 40%–60% in a month, how many runs should an experiment need?
An experiment-design article using Profound, AirOps, and recent measurement research to explain citation drift and a clearly synthetic sample-size example.
Original analysis of archived observations · 2026-09-19
WebMCP status is not platform-neutral
Shopify Liquid accounts for 1,938 of 2,236 native-registration-labelled records in the September snapshot. The platform mix changes how to read the average.
Original analysis of archived observations · 2026-09-19
A WebMCP registration is not an execution
An audit of captured names, raw tool counts, duplicate entries and inactive states in the September WebMCP snapshot.
Operator research · 2026-09-19
What to learn from Sam Blond about GTM in 2026
Five operating decisions from Sam Blond’s public interviews: customer selection, pipeline, first sales hires, AI’s role, and launch distribution.
Practical guide · 2026-09-19
Context Hints in ChatGPT Ads Aren't Keywords
Distinguishes context hints (business facts) from delivery instructions (targeting), with a worked hypothetical brief turned into three hints tagged FACT/ASSUMPTION and a checklist.
Practical guide · 2026-09-19
Why Your ChatGPT Ads Conversions Number Can Move With No Campaign Change
Why identical campaign activity can produce different reports: processing delay, conversion timing and reporting settings, with worked examples.
Practical guide · 2026-09-19
What Your Ad Landing Page Must Expose to OpenAI's Crawlers
A crawler-access diagnostic that separates robots permission, CDN rules, application challenges and actual responses, with a concise engineering handoff.
Practical guide · 2026-09-19
ChatGPT Ad Units vs. Sponsored Agents: What's Live vs. What's Alpha
A dated comparison of standard ChatGPT ads, Sponsored Agents alpha and the separate ad-management features announced in September.
Practical guide · 2026-09-19
A Founder's First ChatGPT Ads Measurement Plan (A Worksheet, Not a Benchmark)
A six-row decision worksheet (conversion event, Pixel/API, attribution window, time basis, observation period, discrepancy triage) demonstrated with a fictional startup's worked example, no invented benchmarks.
Primary-source guide · 2026-09-19
WebMCP vs MCP-B vs server-side MCP: which layer should a founder choose?
A founder decision guide separating the browser WebMCP API, the MCP-B runtime and bridge, and ordinary server-side MCP.
Primary-source guide · 2026-09-19
10 WebMCP resources worth reading
A source-led reading list covering the WebMCP proposal, Chrome's current experiment, MCP-B, server-side MCP, and OpenAI's documented Site Tools surface.
original analysis · 2026-09-19
.com still leads .ai across nine YC cohorts
A current snapshot of listed website TLDs across 1,586 YC directory rows, with .com and .ai counts kept separate from launch-time history or adoption claims.
original analysis · 2026-09-19
YC S26 lists 56 industrial companies, up from 15 in S25
A current YC primary-industry comparison of Summer 2026 against Summer 2025 and Summer 2024, using consistent primary-industry fields and visible cohort denominators.
original analysis · 2026-09-19
Same-origin scanning missed docs links on 202 YC homepages
On the same 1,515 fresh homepage responses, including other-origin links raises docs/developer matches from 188 to 390 companies, while pricing rises from 374 to 381.
original analysis · 2026-09-19
All 70 YC homepages linking llms.txt had meta descriptions
All 70 homepages advertising llms.txt also expose meta descriptions; 67 have canonical links. A fresh, uncapped comparison with ordinary metadata and on-origin versus offsite discovery.
guide · 2026-09-19
How do you diagnose a tool that works behind a Chrome flag but not in an origin trial?
A deployment-first checklist for separating local WebMCP flag success from origin-trial, token, response, and page-policy failures.
guide · 2026-09-19
Which MCP transport should a new remote server choose: stdio, Streamable HTTP, or legacy SSE?
A founder-facing transport decision for local servers, hosted MCP services, and named legacy-client compatibility.
guide · 2026-09-19
How can a CDN or WAF break Streamable HTTP even when JSON-RPC works locally?
A safe edge-debugging runbook for methods, headers, buffering, timeouts, request bodies, Origin checks, and scoped proxy changes.
guide · 2026-09-19
Shopify Cart MCP or Checkout MCP: where should a merchant require credentials?
A documentation-derived commerce guide separating anonymous exploration, checkout preparation, purchase permission, and merchant handoff.
guide · 2026-09-19
Which cross-origin iframe cases require allow=tools and which must fail?
A permission-boundary guide for top-level, same-origin, and cross-origin WebMCP frames, with safe policy and authorization checks.
research article · 2026-09-19
Should a founder treat Google's dedicated generative-AI Search Console report as an impression report, not a citation report?
A current-source guide to what Google’s dedicated generative-AI Search Console report measures, what it does not, and how founders should pair it with behavior and answer evidence.
research article · 2026-09-19
Which ChatGPT Ads attribution window fits a seven-day versus ninety-day buying cycle?
A fit-based attribution decision for short and long buying cycles using OpenAI’s current 7/14/30-day click windows and 0/1-day view-through settings.
guide · 2026-09-19
What makes a product catalog crawlable enough for a feed campaign?
A documentation-derived readiness guide for product data, image hosts, landing pages, robots, WAF/CDN, authentication, rate limits, and feed reporting.
guide · 2026-09-19
How should a directory test canonical, noindex, and redirect changes safely?
A documentation-derived migration guide for choosing redirects, rel=canonical, sitemap inclusion, noindex, and observation steps across a directory URL family.
article · 2026-09-19
When does a first customer require manual service rather than more product?
A founder decision on when manual customer work is useful product discovery, when it becomes hidden agency labor, and when to productize the boundary.
article · 2026-09-19
Should a builder improve the agent harness before changing models?
A failure-first decision guide for choosing between context, tools, state, evaluators, permissions, and a model change.
article · 2026-09-19
Are model price drops a reason to compete on workflow rather than inference cost?
A dated cost snapshot and founder decision on whether falling inference prices should change pricing, reliability, workflow scope, or moat.
company case analysis · 2026-09-19
What can a founder actually copy from Basis's model-routing and reviewability story?
A bounded company-case analysis of routing, reviewable outputs, and workflow-specific evaluation, with vendor claims kept attributed.
curated directory · 2026-09-19
Which primary AI economy datasets can a founder actually reuse?
A practical directory of primary AI-economy datasets and public data products, with dates, access terms, usable questions, and explicit boundaries.
comparison · 2026-09-19
Cranston, Balance and Arden: three different businesses hiding inside AI accounting
A founder-facing comparison of tax-prep software, outsourced AI accounting, and internal-audit automation, with responsibility boundaries and a worked Cranston pricing model.
comparison · 2026-09-19
Prism vs Perfectly: hire a recruiting agency or buy the recruiting workflow?
A recruiting comparison with Prism's dated launch pricing, a worked fee example, and the workflow questions to compare with Perfectly.
comparison · 2026-09-19
Clicks Health, Novoflow and Amera are not selling the same healthcare agent
A healthcare procurement comparison separating provider back-office agents, computer-use EHR deployment, and payer claims infrastructure.
comparison · 2026-09-19
Caseflood vs Concorda: legal AI before the case and during the case
A legal-AI comparison separating intake and client conversion from litigation execution, with Concorda's public Team-plan arithmetic and Caseflood's managed-service boundary.
comparison · 2026-09-19
Logistics is not one product: Getcho, burnt, Pango, or Agent Astra?
A logistics comparison separating delivery reliability software, food-distribution operations, e-commerce post-purchase infrastructure, and operated freight.
comparison · 2026-09-19
Should you buy Crouton or build your own living ownership map?
A founder decision model for choosing a static org chart, an internal allocation layer, or Crouton when customer ownership changes across teams.
analysis · 2026-09-19
What can a CPA firm calculate from Cranston’s per-return pricing?
A dated worked model for comparing Cranston’s public per-book and per-return rates with a CPA firm’s own preparation and review economics.
comparison · 2026-09-19
Should your construction AI start in the design model or the field report?
A construction-market comparison that separates ArchiLabs design automation from Opusense field reporting by buyer, input, deliverable, timing, and price.
comparison · 2026-09-19
Should your agent use Airweave or Crustdata for context?
A founder architecture and buying comparison between Airweave’s private-workspace retrieval and Crustdata’s external people, company, and web data.
analysis · 2026-09-19
What does a founder buy versus configure in Bolna’s voice-agent stack?
A voice-agent stack comparison that separates Bolna’s documented infrastructure from the workflow, handoff, and unit-economics decisions the buyer still owns.
Research and comparison · 2026-09-19
Which YC companies are the most useful public industrial-software peers for a founder?
Six YC industrial-software peers mapped to buyer, deployment boundary and the first proof a founder can test. Includes a dated 56-company homepage-link analysis.
Original research · 2026-09-20
How to measure agent distribution: a 1,752-row WebMCP census
A passive census of 1,752 YC and Speedrun entries found four native registrations in default Chrome and 23 with experimental features. What those counts establish, and what still needs measurement.
Original research and operator guide · 2026-09-21
Use Your Prior Role to Choose a Sharper First Product Test
A screen of 1,586 YC company descriptions and 50 source-reviewed examples shows how prior work can inform a concrete product test, without treating a résumé as proof of demand.
Original research and operator guide · 2026-09-21
A YC directory row is not a prospect until it passes the identity gate
A reproducible nine-cohort YC snapshot shows when a listed website stays on the same host, moves within the same domain, or needs a manual identity check after a cross-domain redirect.
practical-guide · 2026-10-07
How to measure agent traffic without counting the wrong thing
A worksheet for separating crawler requests, page reads, tool use, and verified outcomes.
practical-guide · 2026-10-07
Turn agent questions into useful marketing pages
A worked, fictional example of choosing a page, checking its claims, and measuring what changed.
practical-guide · 2026-10-07
A WebMCP analytics implementation checklist
Verify tool discovery, invocation, failures, and outcomes without counting your own tests as adoption.