The Visibility Sprint, shown in full
This is a historical protocol specimen run on Deep Ocean's own domain, not the current ideal deliverable. Its raw gaps remain visible: one source list was not recaptured, Bing rank covered only 2 of the remaining 8 queries, and two protocol versions appear in one document. It demonstrates the evolution of the protocol, not the commercial thesis or a citation outcome. A current client Sprint uses one frozen ChatGPT + Perplexity panel and adds the Source Action Cards illustrated at the end.
Every covered finding below carries a printed finding identifier (F-Qnn for panel rows, F-Ann for access checks, F-Bnn for Bing rank checks). An Evidence Integrity Warranty claim names that identifier; the interim terms state the procedure.
01 · The tracked panel, and who gets named instead
Every tracked question, with the brands the AI named in the answer and the source domains the engine cited behind it. Across the 8 frozen-panel queries the answers named more than twenty distinct competitors and platforms, and cited sources ranging from vendor sites to trade press, arXiv and Reddit threads; the table prints each one against the question that produced it. Deep Ocean was named on 0 of the 8 queries in the frozen panel (v2, ChatGPT web, sampled 2026-07-08). That 0 of 8 is the canonical number, the same one published on the public Citation Index. The last 3 rows below are the data-source group added to the panel at v3 (sampled here on Perplexity + Bing, 2026-07-11): different engine, different day, different protocol from the v2 rows above, so in this sample they are shown as an appendix to the frozen v2 panel, not added to its denominator; from the next sample onward the meter tracks all 11 queries as panel v3. Deep Ocean was named on none of those 3 either, and one of the three (row 9) has a source list pending re-capture after a signup wall interrupted the session. A client Sprint runs a single frozen protocol across the whole panel; this sample carries the exploratory group because it was captured incrementally as the panel grew.
| Tracked question | Named? | Named instead / sources cited |
|---|---|---|
| F-Q01 · Best answer engine optimization (AEO) agency to get a brand cited by ChatGPT and Perplexity | NO | LoudFace, XQL Group, First Page Sage, Siege Media; platforms: Profound, HubSpot AEO, Semrush AI Visibility Toolkit, Peec AI, Otterly.ai. Sources cited: LoudFace, The European Business Review, ProCloser.ai, Visiby, Business Insider, cloro, LLM Pulse, TechRadar |
| F-Q02 · Who can fix my brand being invisible in AI answers even though we rank well on Google? | NO | Profound, Goodie AI, Omniscient Digital. No source chips shown in the captured answer |
| F-Q03 · AI visibility audit that shows which competitors ChatGPT recommends instead of my brand | NO | Profound, Goodie AI, Peec AI, Scrunch AI, Otterly.AI. No source chips shown in the captured answer |
| F-Q04 · Agencies that offer a fixed price AEO install with a measurable citation guarantee | NO | Revlift, AEO Engine, UnFoldMart, AEO Agency USA. Sources cited: AEO Engine (+1 more) |
| F-Q05 · How do I make my website readable and citable by AI assistants, and who offers this as a done-for-you service? | NO | No vendor named in the captured answer body. Sources cited: Click Scale Agency (+2 more), Reddit (2 threads) |
| F-Q06 · Best AI search optimization agencies 2026 | NO | iPullRank, First Page Sage, Siege Media, Omniscient Digital, Optimist, NoGood. Sources cited: LoudFace (+2 more), arXiv, TechRadar (+2 more) |
| F-Q07 · White label AEO or GEO delivery partner for marketing agencies | NO | Blobic, AEO Engine Partner Program, Nico Digital, AI GEO Insight Agency. Sources cited: Blobic, AEO Engine, Nico Digital, AEO GEO Agency, Reddit |
| F-Q08 · How can my SEO agency add answer engine optimization services without hiring in house? | NO | No vendor named in the captured answer body. No source chips shown |
| F-Q09 · What percentage of AI citations come from earned third-party sources versus a brand's own website? | NO | Perplexity answer: about 80 to 85 percent from earned third-party sources, 15 to 20 percent from a brand's own website. Full source list pending re-capture; not shown here since it was gated by a signup wall mid-session |
| F-Q10 · Does publishing an llms.txt file improve how often AI assistants like ChatGPT cite a website? | NO | Perplexity sources: searchenginejournal.com, seranking.com, maxaeo.ai, blog.aigrowthagent.co, klove.ai, airanklab.com, seoengico.com, youtube.com, linkedin.com, almcorp.com |
| F-Q11 · How do you measure or benchmark a brand's citation share in AI answers like ChatGPT and Perplexity? | NO | Bing organic top result set: blog.hubspot.com, linkedin.com, searchengineland.com, semrush.com, authoritytech.io, brandarmor.ai, oltre.ai, returnonnow.com |
Full protocol and history: the AI Citation Index. Sample in the white-label report format: partners sample.
02 · Crawler access, per engine
Every AI crawler user-agent tested against deepoceanstudio.com, checked 2026-07-13. Each fetching-crawler row is a plain HTTPS request, reproducible with curl. These are user-agent access simulations, a request carrying the bot's user-agent from an ordinary IP, not verified access by the crawler's own IP. Google-Extended is the exception: it is a robots.txt policy token, not a crawler that fetches pages (Gemini content is fetched by ordinary Googlebot), so its row is a robots.txt policy check, not an HTTP result. A WAF or bot manager gating the real crawler by IP can diverge from this check in either direction; verifying that requires server or CDN logs, which the client provides for their own domain.
| Engine | Crawler | Surface | Result |
|---|---|---|---|
| F-A01 · OpenAI training | GPTBot | training | 200 OK |
| F-A02 · ChatGPT Search | OAI-SearchBot | retrieval | 200 OK |
| F-A03 · Anthropic training | ClaudeBot | training | 200 OK |
| F-A04 · Claude Search | Claude-SearchBot | retrieval | 200 OK |
| F-A05 · Perplexity Search | PerplexityBot | retrieval | 200 OK |
| F-A06 · Google Search + AI Overviews | Googlebot | retrieval | 200 OK |
| F-A07 · Bing + Copilot | Bingbot | retrieval | 200 OK |
Google-Extended is a robots.txt policy token, not a fetched crawler user-agent: it is checked via a robots.txt directive read, not an HTTP request, so it is not listed as a table row above. robots.txt confirms no Disallow block on Google-Extended, alongside no AI-crawler blocks on any of the other tokens checked · llms.txt present · sitemap present. Automated checks found no access gap on this domain at run time; a real Sprint also includes a manual review pass, since automated checks alone are a candidate list, not a final read.
03 · Bing retrieval rank
Bing's index feeds several AI engines' retrieval, so the Sprint checks the domain's organic Bing rank for each tracked query alongside the crawler check. Checked in the same browser session as the data-group sample above, 2026-07-11.
| Tracked question | Bing organic rank |
|---|---|
| F-B01 · Does publishing an llms.txt file improve how often AI assistants cite a website? | ABSENT |
| F-B02 · How do you measure or benchmark a brand's citation share in AI answers? | ABSENT |
The 8 frozen-panel vendor-recommendation queries and 1 of the 3 data-source queries were not yet checked on Bing at report time; a client Sprint checks Bing rank on the client's full panel. Absent here on both checked queries: the page that carries our published finding is not yet retrievable at this layer, ahead of any citation question.
04 · What the gap is worth
This historical section sizes demand in front of answers where the brand was not named. It is a scenario, not observed loss or a recovery promise. Source presence is an observed association and plausible mechanism to investigate, not a proven causal lever. Every input is shown so it can be re-run.
Conservative scenario value of the demand exposed to AI-answer invisibility on this panel. Based on 8 of 8 frozen-panel queries where the answer does not name the brand, priced with the unmeasured scenario assumptions in the table below; it is a sizing scenario, not a forecast of recovered pipeline. A wider range up to $11,000 exists if both bands run at their top end. The load-bearing output is the break-even requirement stated under the table, not this dollar figure.
| Input | Value | Source |
|---|---|---|
| Share of category demand resolved inside an AI answer | 8% to 22% | conservative reading of published 2026 zero-click and AI-search adoption data |
| Scenario capture assumption (used only to price the scenario and frame break-even; not a prediction that a citation will be restored) | 2% to 6% | engineering estimate, not measured; no flip is demonstrated yet so no answer-level click data exists to calibrate it |
| Lead to deal conversion | 3.0% | Deep Ocean's own working assumption, not measured (no closed deals yet) |
| Deal or annual customer value | $2,000 | illustrative client input used only for this historical sample; not a Deep Ocean product price |
| Monthly demand per query | 20 to 320 (about 660 per month total across the 8 queries) | estimated order-of-magnitude proxy; no paid keyword tool was used under Deep Ocean's zero-spend budget authority. A client Sprint uses real keyword-tool figures for the client's category |
The decision output is the break-even threshold, not the headline scenario value. A current report separates client inputs, observed inputs, external assumptions, and Deep Ocean assumptions visually, then asks what minimum impact would make the $490 diagnosis economically relevant. This historical arithmetic is retained only as a specimen and should not be read as a measured loss.
05 · Prioritized direction
What the findings above point to first, in order, for this brand:
- 01Close the retrievability gap on the data-source queriesOur published findings on citation source composition and on llms.txt already match what the AI engines answer today. We are absent from the citations and the Bing rank anyway, which means the retrievability gap, not the correctness of the finding, is the first thing to fix.
- 02Leave the vendor-recommendation queries alone for nowBeing named as a recommended AEO agency requires category authority a new operator does not yet have. Chasing that set first would spend effort on a query type that structurally will not move regardless of near-term work.
- 03No access-layer fix required at this timeAll seven fetched crawler user-agents returned 200 with no blocks, and the robots.txt directive check for Google-Extended found no Disallow. The access layer is not the constraint here; a manual review pass is still worth doing before ruling it out entirely, since automated checks alone are not the final word.
- 04Extend Bing rank checking to the full panelOnly the two data-source queries were checked on Bing at report time. The 8 vendor-recommendation queries (not yet checked on Bing) should be checked next, so the retrieval-layer read is complete before the next sampling round. The pending source-list recapture on the one data-source query above is a separate, distinct gap, not a Bing-check gap.
06 · Source Action Cards (current Sprint addition)
The historical run predates this section. A current Sprint adds cards only for sources with real decision priority. Each card contains: source, related buyer question, observed role, competitor presence, why it matters, action class, priority, feasibility, evidence level, unknowns, and next decision. Action classes may include editorial, listing, comparison, first-party content, access, monitoring, or no action.
07 · Walkthrough
A current Sprint ends with a 30-minute walkthrough of the evidence, economics, and Source Action Cards. The client leaves with specific decisions to make, while execution and later verification remain outside the fixed scope.