August 19, 2026 — retrieval architecture and matched discovery
Dataset v1.421, a clean firewall audit and exact-address controls
Quoted First published version 1.421 of its ChatGPT-native search observation dataset. The release contains 2,195 dated JSON observations and controls through 07:01 UTC, plus 2,626 CSV records. It adds exact revision 950 production integrity, native and direct-open negatives, a clean project-level Vercel firewall audit and the first verified crawler event for that deployment.
Verified GPTBot fetched an earlier comparison revision twice at 05:10 and 05:15 UTC on August 18. Roughly 24 hours later, the generic query, historical and current headlines, title, brand, target-domain and literal-URL checks still omitted Quoted First. The current body changed after the GPTBot sweep and had no verified GPTBot, OAI-SearchBot or Bing request to the path in its checked window, so the result is not presented as an unchanged-current-body test. The full result is in Does a GPTBot crawl mean a page is indexed in ChatGPT search?
The follow-up supplied that article's complete canonical URL in a native prompt and asked for a summary. The search rewrote around the topic and did not return or cite the page; exact Vercel metrics recorded no verified bot request in the completed 05:30–05:45 UTC window. OpenAI's official link-safety documentation supplies the missing mechanism: automatic agent retrieval can check whether an exact URL was previously observed by an independent web index. The stopped Quoted First opens are consistent with that gate but do not prove which internal path handled one request. Unrelated accounting controls showed that architecture and workflow constraints shifted the admitted candidate mix while a $40 cap remained soft.
Four-hour matched IndexNow test
The matched experiment uses two frozen Vercel sites with the same visible answer, price, schema, crawler rules and native-search queries. Only one received an IndexNow notification. In the completed 02:30–06:31 UTC window, verified Bing requested only that arm's ownership key, verified OAI-SearchBot requested only the untreated arm's robots.txt, and ClaudeBot fetched both roots and both robots files. Both original deployments retained their exact bodies, ETags and checksums. Neither site appeared for the natural buyer query, exact title, token, hostname, exact-domain or direct-open tests. A broader vercel.app restriction did return relevant six-day-old law-firm pages, so the hosting family was eligible. The full design and limitations are in Does IndexNow get a new Vercel site into ChatGPT search?
Search architecture controls
OpenAI's public documentation, plus comparison research into Exa and Firecrawl, motivated a layered model: URL discovery or corpus admission comes before retrieval and reranking, while result snippets and full page extraction can be separate states. Quoted First does not claim that Exa or Firecrawl powers OpenAI. Unrelated admitted-domain controls succeeded by domain restriction and direct open while equivalent Quoted First controls stopped before a target fetch.
The new answer How does ChatGPT web search find and open a new URL? turns those findings into one focused informational document. It separates prompt fan-out, candidate retrieval, exact-address safety, extraction and citation; publishes the Presenc, BlueprintAI and Google Play controls; and explains why scenario specificity helps only after a host is admitted. This is a source-backed model, not disclosure of a private ranking algorithm.
The revision 950 read-only firewall snapshot reported no active WAF configuration, custom rules, IP blocks, system bypasses or drafts, with Attack Mode off. Public parity, cache HIT, conditional 304 and error-log checks passed. Verified metrics showed one ClaudeBot request to /robots.txt at 06:59 UTC but no verified article or OpenAI request. Natural, title, host-restricted, URL, recency, distinctive-fact, direct-open and generic native checks still omitted Quoted First.
August 18, 2026 — exact identity, buyer scenarios and frozen formats
Visible identity and disambiguation
The site added What is quotedfirst.com?, a visible factsheet mapping Quoted First, QuotedFirst, QuotedFirst.com and quotedfirst.com to one managed LLM search optimization service. It explicitly distinguishes unrelated quotation, insurance and similarly named software businesses. The intervention followed a native-search control in which the unquoted domain token resolved toward QuoteFirst.ai and the exact quoted domain returned no target.
Specific buyer questions
New answer pages addressed small personal-injury firms, lean B2B SaaS teams, pre-revenue companies and buyers seeking month-to-month managed implementation below $3,000. The pages disclose the same first-party offer facts: Starter is $2,900 per month, tracks 150 named prompts weekly across six engines and includes implementation support. They also disclose zero customers and no placement guarantee.
Independent format and hostname controls
Separate, frozen Vercel projects tested HTML, plain text, Markdown, JSON-LD and PDF roots without modifying older controls. Verified third-party crawler delivery occurred on several projects, while OAI-SearchBot activity remained limited to policy files in the completed windows. The formats and keyword hostnames did not create immediate ChatGPT-native candidate admission.
August 17, 2026 — open method and vertical retrieval studies
Open measurement specification
Quoted First published its AI search measurement methodology: 150 named prompts across six engines, 900 scheduled prompt-engine checks per weekly collection, fixed collection rules, formulas, missing-data policy and limitations. The method is an operating specification, not evidence of customer performance.
Personal injury and B2B SaaS samples
Two 30-result studies preserved the first ten URLs for three personal-injury prompts and three B2B SaaS provider prompts in the OpenAI web-search tool available to the research environment. The personal-injury sample showed scenario wording could favor relevant law-firm pages; the B2B SaaS sample showed provider-authored comparisons dominated the tested query family. Both are small point-in-time samples, not disclosures of a ranking formula.
August 16, 2026 — crawler eligibility and public evidence
The site published a crawler checklist, canonical sitemaps, Atom, RSS and JSON feeds, llms.txt, plain-text and Markdown alternates, structured organization facts and the first versioned native-search observation dataset. Public responses allowed OAI-SearchBot, ChatGPT-User and other named search crawlers. Server evidence began separating a public HTTP 200, crawler policy retrieval, verified content delivery and native candidate selection as different events.
August 15, 2026 — initial public website
Quoted First's initial public website described a managed LLM search optimization service, transparent monthly pricing, a prompt-measurement workflow and buyer education about source selection. The commercial core has remained intact while later releases added evidence, specific answers and controlled discovery surfaces around it.
What qualifies for the launch log?
An entry must describe a substantive public artifact, controlled test, correction or operational disclosure. Repeated deployments with no meaningful content or method change do not receive standalone entries. Each experiment entry distinguishes:
- a successful deployment from a successful public response;
- a discovery notification from a verified crawler request;
- a policy-file request from a content request;
- candidate inclusion from ranking and citation; and
- first-party claims from independent evidence.
Deployment identifiers, query wording, timestamps, checksums, crawler filters and negative results live in the full experiment record. This shorter page exists so a reader or retrieval system can understand the publisher's chronology without parsing the entire dataset.
Frequently asked questions
Is this a customer-results page?
No. Quoted First currently reports zero customers. The log covers the company's own website, research method and controlled search experiments.
Does a launch log make a new site rank in ChatGPT search?
No. It supplies crawlable dates, identity and first-party evidence. It does not force a crawler request, corpus inclusion, ranking, recommendation or citation.
Why publish negative results?
Because a public page, accepted submission or crawler request is often misreported as proof of indexing. Preserving failed checks makes later timing claims auditable and prevents the method from selecting only successful observations.