The Optiview Experiment

Can clearer information change what AI says?

We help businesses publish clearer website information and approved details about what they offer. Then we examine what AI services receive, what appears in their answers, and what we still cannot explain.

How we test →

Perplexity × Rhythm90

A 17-word match between a published source and an AI answer.

The question did not mention Rhythm90. The same 17-word passage appeared in its published business context and in the recorded Perplexity answer.

Published Narrative Graph · excerpt
an illustrative teaching exercise, not a customer case study; its people, counts, and source identifiers are fictional
Perplexity answer · excerpt
an illustrative teaching exercise, not a customer case study; its people, counts, and source identifiers are fictional

The words match. We cannot establish how Perplexity obtained them or whether the Narrative Graph influenced its recommendation.

Read the question and full answer →

More to investigate

Readable content is only part of the journey.

Four services kept the important qualifications.

In an October 31 test, Exa, You.com, Jina and Firecrawl returned key About-page facts, including our no-guarantee qualification. Links and introductory text survived differently. None returned the selected graph-only markers.

Compare what each service returned →

A service returning text, a website serving content and an AI answer using that content are separate observations.

Explore AI agent coverage and delivery rules →

Four observations. Four different questions.Evidence framework
  1. 01Prepared source

    What did we publish?

    ?
  2. 02Delivered response

    What did the website serve?

    ?
  3. 03Retrieved text

    What did an external tool return?

    ?
  4. 04AI answer

    What appeared in the answer?

Each link needs evidence. Observing one step does not prove the next.

A conceptual framework, not a measured conversion funnel. Actual observations and their limitations are documented separately.

Earlier observations: wording, qualifications and retrieval

What we’re learning

Three things the evidence tells us.

AI answers sometimes keep important qualifications.

Some answers correctly described service limits and other important qualifications. We have not shown that Optiview improves answer accuracy overall.

See the examples →

Publishing information does not guarantee retrieval.

In four direct responses from Exa and You.com, we did not find the complete passage we had added to the published source. Some related information appeared, but not the full passage.

Read the retrieval study →
One website. A clearer way to read it.

The idea behind the experiments

One website. A clearer way to read it.

Visitors keep seeing the normal website. Eligible AI requests can receive a simpler version of its content, plus any business information the team has approved. The optional Narrative Graph gives that information structure, including what the business offers, who it serves, and important limits.

The business does not have to maintain a second public website for this conversion.

Six questions our experiments are designed to answer.

The six questions we’re trying to answer

Six questions our experiments are designed to answer.

Question 1

Will AI find your business?

When someone asks for a product or service without naming a company, does an AI answer include that business? Our retained tests include both appearances and omissions. We have not shown that Optiview increases the likelihood of inclusion.

See a local-service example →

Question 2

Will AI use the information you publish?

Some recorded answers repeat distinctive published wording; others express related ideas in different words. In many cases, we cannot identify the particular source or retrieval path behind the answer.

Compare the words →

Question 3

Will AI understand what makes you different?

We look for correct descriptions of a business’s purpose, offerings and limits. Some answers preserve those distinctions, but we cannot isolate the effect of the additional content we published.

See an example →

Question 4

Will AI get the details right?

We check details that could change a customer’s decision: prices, who provides a service, what is included and whether an example is real. Some answers got these details right. We have not established a consistent improvement in accuracy.

Read the answer →

Question 5

Can we tell what AI actually received?

We can record what Optiview served in an observed website response and compare that with results from outside retrieval services. Neither tells us, on its own, exactly what information reached the model generating an answer.

See a matching website delivery record →

Question 6

Does this make a measurable difference?

We want to know whether clearer published information changes how often a business appears or how accurately it is described. Our version-comparison tests did not produce reliable enough automated scores to establish an improvement.

See how we tested this →
Explore the retained AI conversations

See the experiments for yourself

Read the actual questions and recorded AI answers.

Open any record to compare the question, the captured answer, the published information and the supporting evidence. Some questions supplied a website address; others asked about a need without naming the business. Each record identifies the test type.

These examples are selected from the retained tests. The complete results also include answers that did not mention the business.

Does clearer information lead to better answers?

What remains unanswered

Does clearer information lead to better answers?

Our retained tests show businesses appearing in some AI answers, passages of published language being repeated, and important qualifications being preserved. They do not show a consistent improvement caused by Optiview.

We completed the direct Exa and You.com retrieval checks. To test impact more rigorously, a future study would need to compare answers with and without approved context under controlled, verified retrieval conditions. That would help distinguish an observed match from a measured effect.

Read the studies and their limitations →
Help businesses make their meaning available.

The bigger idea

Help businesses make their meaning available.

AI systems do more than find information: they interpret what they find and assemble answers. We’re testing whether clearer, business-approved information can help those interpretations—and publishing results that support the idea as well as results that challenge it.

One study examines OpenAI’s published gpt-oss browsing-tool code to understand what a browsing tool can retrieve. The study is run by Optiview. It is not endorsed by OpenAI and does not test production ChatGPT browsing.

Research methods and complete results

View the original evidence

How we check website visits and deliveries