Evidence & methodology

See published brand context appear in live AI answers

Inspect real conversation excerpts, matching delivery records and retained source files. See what the evidence establishes, from retrieval to quoted wording.

Observed in live AI conversations

Published context. Retrieved source. Words in the answer.

Nine documented conversations across ChatGPT, Claude and Perplexity, using four Optiview-powered sites. Each pairs answer excerpts with a production delivery whose SHA-256 matches the retained response body.

Tested October 22, 2026, US Eastern. Two unbranded prompts and seven directed URL tests. First-party evaluation by Optiview; selected examples, not a benchmark or a guarantee.

9distinct conversations
4published websites
2unbranded prompts
01 · ClaudeUnbranded · search explicitly requested

Found without naming Optiview

Optiview ↗

From the captured answer

It says it does not control an LLM’s answers or guarantee placement, and that each AI service decides what to retrieve, use and cite.
Delivery reports show what was served, not whether it appeared in an AI answer.

Distinctive graph wording reused inside a comparison with other vendors. No quotes or target brand were requested.

Markdown deliveredgraph:19 · 2026-10-23 03:00:19 UTC
Captured Claude answer using published context about Optiview
Historical conversation capture. Wording and product interface preserved as observed. Open for full size.
Inspect prompt, comparison and delivery receipt

Exact prompt

Please search current web sources for services that publish reviewed brand context alongside structured website content for AI assistants. Compare their connection methods, publication controls, and promises about AI citations.

What the comparison establishes

Matched graph wording was absent from the prompt and captured cited original pages. A response-body hash matched a production delivery during the test window. Timing supports correlation; it does not cryptographically bind a request to a conversation.

Recorded requester
Claude-User
Identity evidence
Declared User-Agent; no independent provider IP verification
Delivered body SHA-256
3ef15226e62a1f889152a57a396d208f0d882851c6e12a9c341b0108e9db17b9

Temporary/incognito conversation; no persistent public conversation link. Prompt and answer excerpts retained here.

02 · PerplexityUnbranded · specific product need

An unbranded need finds Optiview

Optiview ↗

From the captured answer

Information can apply across the website or to one exact page path.

An eight-word graph match inside a broader explanation. Short phrase: weaker evidence of graph-specific use than the longer examples.

Prepared source delivered as HTMLgraph:19 · 2026-10-23 02:54:59 UTC
Inspect prompt, comparison and delivery receipt

Exact prompt

Is there a service that turns website pages into structured Markdown and combines them with team-approved brand information? How does it connect to a site, and does it guarantee AI citations?

What the comparison establishes

Matched graph wording was absent from the prompt and captured cited original pages. A response-body hash matched a production delivery during the test window. Timing supports correlation; it does not cryptographically bind a request to a conversation.

Recorded requester
PerplexityBot
Identity evidence
Provider IP check matched
Delivered body SHA-256
2c390c8442ec102a0dcb90dc3bc0124d694aca72311724b70775f8386c9c13f0

Temporary/incognito conversation; no persistent public conversation link. Prompt and answer excerpts retained here.

03 · ChatGPTDirected URL test

Who actually performs the repair?

A5 Home Services ↗

From the captured answer

A5 is a referral and matching service; the independent provider assesses the fixture and performs any agreed work.
No charge to submit a request or use A5's matching service. Contractors charge separately for their work.

Quoted graph wording; requested short supporting quotes.

Prepared source delivered as HTMLgraph:15 · 2026-10-23 02:19:01 UTC
Captured ChatGPT answer using published context about A5 Home Services
Historical conversation capture. Wording and product interface preserved as observed. Open for full size.
Inspect prompt, comparison and delivery receipt

Exact prompt

Retrieve https://www.a5homeservices.com/services/plumbing/running-toilet live. Based only on that page, explain who performs the repair, what homeowner matching costs, whether A5 introduces several providers at once, whether a match is guaranteed, and what credential checks homeowners should make. Give source citations and short exact supporting quotes; state any unknowns.

What the comparison establishes

These passages were present in the published graph and prepared Markdown, and absent from the captured original page. This comparison does not cover every other page or search cache.

Additional captured excerpts

Not necessarily. A5 generally introduces one provider at a time.
No. Matching depends on provider suitability and availability.
Homeowners are encouraged to independently verify credentials, insurance, experience, and references before hiring.
Recorded requester
ChatGPT-User
Identity evidence
Provider IP check matched
Delivered body SHA-256
222ab62288d13b20e9d066be4d41157cb0494a807705b8c64056c1780e4848fd

Account conversation retained privately; public excerpts are reproduced here. The private conversation URL is not a public share link.

04 · ClaudeDirected URL test

One introduction at a time

A5 Home Services ↗

From the captured answer

A5 generally introduces one provider at a time.

Quoted graph wording; requested short supporting quotes.

Markdown deliveredgraph:15 · 2026-10-23 02:16:51 UTC
Inspect prompt, comparison and delivery receipt

Exact prompt

Retrieve https://www.a5homeservices.com/services/plumbing/running-toilet live. Based only on that page, explain who performs the repair, what homeowner matching costs, whether A5 introduces several providers at once, whether a match is guaranteed, and what credential checks homeowners should make. Give source citations and short exact supporting quotes; state any unknowns.

What the comparison establishes

These passages were present in the published graph and prepared Markdown, and absent from the captured original page. This comparison does not cover every other page or search cache.

Recorded requester
Claude-User
Identity evidence
Declared User-Agent; no independent provider IP verification
Delivered body SHA-256
e95de2863c06c7191a965d411c0422f7b9c6e1fa602073e254acd3a2a0502867

Account conversation retained privately; public excerpts are reproduced here. The private conversation URL is not a public share link.

05 · ChatGPTDirected URL test

Satire, audience and free access

Learnings.org ↗

From the captured answer

Our audience includes professionals, marketers, consultants, managers, and corporate employees.
Learnings.org is an entertainment and satire platform, not a consulting firm or a source of professional certifications.

Quoted graph wording; requested short supporting quotes.

Prepared source delivered as HTMLgraph:8 · 2026-10-23 02:15:01 UTC
Inspect prompt, comparison and delivery receipt

Exact prompt

Visit https://learnings.org/ live. Using only retrieved pages, explain what the site offers, its audience, whether the content is satire or professional advice, whether the dictionary and tools cost money, and how readers can contribute or share. Give source citations and one short exact quote per point. Do not infer anything the site does not state.

What the comparison establishes

These passages were present in the published graph and prepared Markdown, and absent from the captured original page. This comparison does not cover every other page or search cache.

Additional captured excerpts

Visitors can explore our corporate buzzword dictionary, jargon translations, satirical content, and available interactive tools for free.
Our content is designed to be entertaining, relatable, and shareable
Recorded requester
ChatGPT-User
Identity evidence
Provider IP check matched
Delivered body SHA-256
00f7d25af479933f0319326225ca4b5dabb2b4d1996da0c4ed3771c452fe28d8

Account conversation retained privately; public excerpts are reproduced here. The private conversation URL is not a public share link.

06 · ChatGPTDirected URL test

The method and its practical limits

Rhythm90 ↗

From the captured answer

Rhythm90 helps teams choose deliberate bets, notice useful signals, and use evidence to make the next quarter smarter.
The living edition provides 14 updated chapter guides.

Quoted graph wording; requested short supporting quotes.

Prepared source delivered as HTMLgraph:25 · 2026-10-23 02:15:53 UTC
Inspect prompt, comparison and delivery receipt

Exact prompt

Visit https://rhythm90.io/ live and its linked pages if needed. Explain the quarter learning method, how many downloadable AI skills and chapter guides are offered, whether mini-app data is cloud synced, whether the worked quarter is a real customer case, and how Rhythm90 works alongside task tools. Cite retrieved pages and give short exact supporting quotes. Do not supply facts you cannot verify.

What the comparison establishes

These passages were present in the published graph and prepared Markdown, and absent from the captured original page. This comparison does not cover every other page or search cache.

Additional captured excerpts

They do not automatically sync to other browsers or teammates.
Recorded requester
ChatGPT-User
Identity evidence
Provider IP check matched
Delivered body SHA-256
0b7bf02db8f85ac61af5d18b90df3120596146fd825c3d0046d85baebff2b37d

Account conversation retained privately; public excerpts are reproduced here. The private conversation URL is not a public share link.

07 · ClaudeDirected URL test

A fictional example stays fictional

Rhythm90 ↗

From the captured answer

an illustrative teaching exercise, not a customer case study.
They do not automatically sync to other browsers or teammates.

Quoted graph wording; requested short supporting quotes.

Markdown deliveredgraph:25 · 2026-10-23 02:15:09 UTC
Inspect prompt, comparison and delivery receipt

Exact prompt

Visit https://rhythm90.io/ live and its linked pages if needed. Explain the quarter learning method, how many downloadable AI skills and chapter guides are offered, whether mini-app data is cloud synced, whether the worked quarter is a real customer case, and how Rhythm90 works alongside task tools. Cite retrieved pages and give short exact supporting quotes. Do not supply facts you cannot verify.

What the comparison establishes

These passages were present in the published graph and prepared Markdown, and absent from the captured original page. This comparison does not cover every other page or search cache.

Additional captured excerpts

one to three Plays, or one for a first pilot.
The living edition provides 14 updated chapter guides.
Rhythm90 does not replace Agile, Asana, or a campaign calendar.
Recorded requester
Claude-User
Identity evidence
Declared User-Agent; no independent provider IP verification
Delivered body SHA-256
af0b1e93de18988066131aeefdbcd8e672634126e0e7db01e0e1b8914c520860

Account conversation retained privately; public excerpts are reproduced here. The private conversation URL is not a public share link.

08 · ClaudeDirected URL test

No promise of AI placement

Optiview ↗

From the captured answer

Optiview does not control an LLM’s answers or guarantee placement in them.

Quoted graph wording; requested short supporting quotes.

Markdown deliveredgraph:19 · 2026-10-23 02:15:55 UTC
Inspect prompt, comparison and delivery receipt

Exact prompt

Visit https://optiview.ai/ live and linked product pages if needed. Explain how the published narrative combines with parsed website source, who controls approval and publication, how it differs from an AI visibility tracker, which connection methods it offers, and what it promises about retrieval and citations. Cite retrieved pages and include short exact supporting quotes. Do not infer guarantees.

What the comparison establishes

These passages were present in the published graph and prepared Markdown, and absent from the captured original page. This comparison does not cover every other page or search cache.

Recorded requester
Claude-User
Identity evidence
Declared User-Agent; no independent provider IP verification
Delivered body SHA-256
3ef15226e62a1f889152a57a396d208f0d882851c6e12a9c341b0108e9db17b9

Account conversation retained privately; public excerpts are reproduced here. The private conversation URL is not a public share link.

09 · ChatGPTDirected URL test

Website source plus published brand context

Optiview ↗

From the captured answer

Optiview turns complex website content into clean, structured Markdown and combines it with the brand narrative your team chooses to publish.

Quoted graph wording; requested short supporting quotes.

Prepared source delivered as HTMLgraph:19 · 2026-10-23 02:17:39 UTC
Inspect prompt, comparison and delivery receipt

Exact prompt

Visit https://optiview.ai/ live and linked product pages if needed. Explain how the published narrative combines with parsed website source, who controls approval and publication, how it differs from an AI visibility tracker, which connection methods it offers, and what it promises about retrieval and citations. Cite retrieved pages and include short exact supporting quotes. Do not infer guarantees.

What the comparison establishes

These passages were present in the published graph and prepared Markdown, and absent from the captured original page. This comparison does not cover every other page or search cache.

Recorded requester
ChatGPT-User
Identity evidence
Provider IP check matched
Delivered body SHA-256
1cf67782d4766c53c517408f491ee87e46af5241cd28362af7ac58470aa48666

Account conversation retained privately; public excerpts are reproduced here. The private conversation URL is not a public share link.

Are the models reading it, or are we steering the answer?

The two unbranded prompts supplied no site name, URL or graph wording. Claude found Optiview in a vendor comparison; Perplexity found it from a specific product need. The other seven prompts supplied a URL and requested short supporting quotes. They demonstrate retrieval and use under directed testing, rather than spontaneous discovery.

The answers mix paraphrase with retained wording. The quoted graph passages were absent from the captured original pages. That supports use of the published additions; it does not reveal the model’s internal reasoning or exclude every possible cached source.

Read the method, wider results and limits

Nine selected conversations, not nine independent replications of one benchmark. Seven directed URL tests plus two unbranded tests. No graph wording was pasted into these prompts. Directed tests requested exact supporting quotes.

Four production sites with published Narrative Graphs: A5 Home Services, Learnings.org, Rhythm90 and Optiview. We operate these sites. Signed-in product accounts and temporary/incognito conversations are not a guarantee of an anonymous or unbiased ranking test.

SHA-256 confirms retained response bytes match the production delivery record. Provider-IP status is separate from User-Agent. Timestamp and source/answer overlap support retrieval correlation, not per-conversation cryptographic attribution or proof of a model’s internal reasoning.

Models summarize and paraphrase, and sometimes retain exact phrases. Quote matching normalizes case, whitespace and punctuation; it does not imply that the entire answer is copied.

The broader unbranded study retained 64 prompt/product combinations (66 attempts after two error retries): 63 usable answers, 19 target-site mentions, 44 target omissions and one inconclusive timeout. Five conversations matched graph language; two also had fresh hash-matched delivery. Gemini supplied no qualifying example. These counts are descriptive, not an adoption or success rate.

Selected positives are not typical-result, ranking, revenue, indexing or citation guarantees. HTML examples show prepared source plus graph delivered as HTML; they do not establish text/markdown on the wire. Public files omit account history, customer IDs, IP addresses and raw internal telemetry.

Verify a retained body locally with shasum -a 256 filename.txt, then compare the result with the case receipt. These are publisher-retained records, not an independent third-party attestation.

Download all nine case receipts and methodology (.json)

What Optiview can verify

Source content, saved changes, publication versions and delivered responses. Your organization manages factual approval.

What you measure separately

AI answer accuracy, citations, brand mentions and downstream business impact. Before-and-after changes alone do not establish causation.

Useful information, with its source

The website preview cleans and organizes the submitted page’s content, headings and links. Inspect the readable result alongside the exact Markdown. It is not a whole-site scan or a completeness score. A payment logo does not establish checkout eligibility, and a returns link does not establish its policy.

A source you can inspect

Know what you’re putting into the world

Follow your content from the received page to the response Optiview delivers.

01

Source

Compare the received page with its structured content. Inspect retained facts, links and any limitations.

02

Reviewed content

Review the brand details, answers and supporting context your team prepares. Your organization owns factual approval.

03

Publication

Identify the content version made live. Saving a draft is separate from publishing.

04

Delivery

Check which representation and publication version Optiview served through the configured connection.

A published version records what went live. It does not prove that a separate reviewer signed off.

Measure the next step

A clearer source is the starting point

Optiview gives you control over what you publish. AI services decide what they retrieve, use and cite. A clearer source does not guarantee inclusion or better outcomes.

How publishing and measurement work together
Inspect in Optiview

Preparation, publication and delivery

Check the content, version and response associated with your page.

Evaluate with your own tools

AI answers, citations and business results

Monitor observed changes over time. A delivery record alone does not establish AI retrieval, indexing, use or citation.

A practical evaluation

Start with a question. Keep the comparison fair

Use Optiview alongside your existing AI visibility and analytics tools.

  1. 01

    Set a baseline

    Choose the questions, reference facts and business measures that matter. Record your starting observations.

  2. 02

    Document the change

    Review and publish the content you want to improve. Verify the delivered response and keep the page, version and evaluation dates together.

  3. 03

    Observe what follows

    Track answers, citations and relevant visits or inquiries. Compare repeated observations, not a single response.

Models, indexes and other sources also change. Before-and-after movement alone does not establish that Optiview caused it.

For a closer look

The checks behind the source

Our internal validation tests known extraction and composition behavior. It is not an independent benchmark of improved AI answers or complete extraction on every website.

Parser validation resultsRecorded tests, versions and scope

REVIEWED EVIDENCE: October 22, 2026

Single-page parser: semantic-v3.10.14. The deterministic parser includes source-owned title and editorial metadata repairs, responsive heading handling, interface heading exclusion, punctuation serialization, source-backed appendix labels, and valid Markdown for linked cards. Core parsing does not generate FAQs, infer company-wide information or use an LLM. Existing source-review bindings, custom selectors and approved published supplements retain their review controls.

Parser validation

An internal engineering check of the established parser on a fixed corpus. It checks literal code preservation and faithful composition, not improved AI answers or complete extraction on every website.

Source executions
660
Unexplained differences
0
Composed outputs
65
Unchanged extraction abstentions
21

660 frozen-source serializer comparisons under the repaired parser, plus exact output comparisons on 1,000 core sources and 660 retained inputs against the accepted engineering candidate. Generated labels are distinguished from source prose. These retained-source checks do not establish 99.9% accuracy or AI outcomes.

Extraction semantic-v3.10.14Composition composition-v2.1.0
Corpus, checks and scope

Corpus: 164 gold inputs, 378 mutations, 18 meaningful controls and 100 captured real pages.

  • Serializer equivalence under the repaired parser, with separate frozen-output comparisons against the accepted engineering candidate.
  • Literal code payload preservation, source-prose checks and source inventory comparison.
  • Destination preservation and deterministic composition.
  • Adversarial cases covering distinct topics, unequal memberships and ambiguous directory spans.

These versions identify the source-code baseline reviewed for this site. For an individual result, inspect its returned versions and source mode. A static website label does not attest to every deployed environment.

How content is prepared and reviewedSoftware, AI suggestions and your team’s role

Software structures received content

Extraction and composition organize source content without a generative model. They do not create new descriptions or decide which claims are true.

AI suggestions are a separate step

The core parser does not generate FAQs, infer company-wide information or use an LLM. FAQ text already on the submitted page remains source content. Narrative Graph is a separate publishing workflow with its own review controls. The public demo can generate illustrative AI drafts; admin AI Assist is currently unavailable.

Your team checks facts and reviews any additions before publishing. The public preview neither saves nor publishes content.

How reviewed content enters publishing
Extraction and composition: two separate checks

Extraction records source content, structure and link occurrences. Composition presents equivalent directory groups once and organizes source-owned headings, offerings, promotions, customer information and attributed reviews when their relationships can be verified.

  • Extraction and composition use no generated descriptions, AI importance ranking or cross-page fetching.
  • Distinct groups and unequal memberships remain separate. Original wording, ordered destinations and occurrence evidence remain available; the organized view can reorder complete source-owned groups with explicit mappings.
  • A separately versioned presentation pass adds separators to generated link labels only where the same source HTML establishes block boundaries. A separate readable presentation can join source-owned title lines, label image descriptions, separate utility context and omit redundant labels for the same source link. Independently matched standalone interface controls can be suppressed. Original labels, controls and Markdown remain available in evidence.
  • Ambiguous cases retain the original representation.

Compare the received page, original compiler output and composed source. Evidence-enabled previews include original Markdown and the occurrence inventory. The demo inspects one page without generating an answer or acquiring linked pages for enrichment.

A fact missing from the original output is an extraction issue. Poor presentation of retained information is a composition issue. Preserved inventory does not mean every element of an arbitrary website was extracted.

Source checks and exact-source approval

Source checks can pass, require review or report an unavailable source. Warnings may identify unresolved relationships, competing states or incomplete content. Passing checks does not certify that the website supplied every fact.

The public demo may show useful received content as a qualified preview when source checks require review. This is inspection only: it does not approve the source for publishing. Holding pages and unusable responses remain outside the transformation view.

Publishing brand content does not approve a flagged source. An authorized user must review the published-version preview and approve that exact page source. Relevant source, settings or version changes require another approval.

Saved additions and published versions

Inspect the approved Narrative Graph additions and compatible historical content that match each page, together with the publication version used for delivery. Your organization manages factual approval; a published version does not prove a separate reviewer signed off.

Connection and delivery checksWhat a response can—and cannot—tell you

Optiview can fetch a supported public URL, use configured browser rendering, or process authorized HTML supplied through an API integration. Source checks look for known error responses, access challenges and incomplete page shells; automated validation does not catch every failure.

Applicable approved Narrative Graph additions are combined with the matching page; historical publications remain compatible. Optiview serves Markdown to requests that explicitly prefer text/markdown. Recognized AI tool names and supported Cloudflare bot categories can also qualify: AI tools with absent or wildcard Accept receive Markdown, while requests accepting HTML can receive the same prepared source and applicable published graph as HTML. Traditional search crawlers, including Googlebot, bypass Optiview conversion and receive the original site response, even if they request Markdown. A declared tool name permits public content delivery but does not verify identity. Published provider IP ranges and Cloudflare category evidence are reported separately. Ordinary browsers retain the original page. Connection, entitlement, source-review and publication checks still apply; authenticated, cookie-bearing and range requests are not converted. Delivery does not guarantee indexing, AI use or citations. API integrations use their own authorized delivery path.

Supplying authorized HTML can avoid an external fetch. It does not execute missing client-side content, guarantee extraction quality or remove security requirements.

Requester identity needs separate evidence

A User-Agent label or preference for Markdown is not verified identity or a crawler adoption rate. Report independently verified, unverified and test requests separately. A manual request tests delivery; it is not evidence of crawler adoption.

Future benchmarksProposed evaluations, not published results

Source retention

Planned: establish an independently reviewed reference set, then measure retained, omitted or changed facts, links, table relationships and qualifications, along with unsupported additions. Evaluate initial HTML and browser-rendered extraction separately, reporting settings and failures alongside results.

Controlled answer quality

Planned: compare the original source with Optiview output using the same questions, model, prompt and settings. Separate extraction-only output from output with approved additions. Use repeated trials and blind scoring against reference facts to measure correct answers, unsupported assertions and unanswered questions.

These independent benchmarks remain planned. Internal regression tests do not establish universal extraction quality, and a controlled answer-quality result would not establish real-world crawling, indexing or citation improvements.

See for yourself

Start with your own website

Inspect a preview, or talk with us about evaluating a supported setup.