> ## Documentation Index > Fetch the complete documentation index at: https://docs.openlens.com/llms.txt > Use this file to discover all available pages before exploring further. # Perplexity website and API: how results compare > Similar guidance, variation in the details, and what the comparison means for your OpenLens reports. Perplexity's API helps OpenLens keep your tracking running when website collection is unavailable. It provides AI answers, brand mentions, and cited sources for your visibility reports. We compared eight questions using saved website answers and new API answers. **The overall guidance was broadly similar in our reading of the answers.** Both sources covered familiar approaches to AI visibility: tracking relevant buyer questions, comparing competitors, reviewing citations, and reporting changes over time. Wording, named tools, and source pages varied within those shared themes. The website answers were collected on July 1–2, 2026, and the API answers on September 17, 2026. Normal variation between AI answers and this 77-day gap both provide useful context for understanding the results. ## What we found | Finding | What we observed | | --------------------------------- | --------------------------------------------------------------------------------------------------------------------------- | | Broadly aligned guidance | Both sources addressed the same practical goals and often recommended similar workflows and evaluation criteria. | | Different levels of detail | Website answers often explained a general approach; API answers more often added named tools and specific examples. | | Variation across repeated answers | Asking the API the same question twice produced changes in phrasing, tool lists, and citations, alongside recurring themes. | | Completed answers with sources | All 16 API requests returned an answer with citations. | The assessment of shared meaning comes from reading the answers. The numerical comparisons below measure exact brand names, source pages, and answer length; a formal semantic-similarity score would require a separate assessment. ## Shared themes in the answers These examples illustrate where the substance aligned. Question descriptions are shortened; the [study data](#explore-the-study-data) contains the exact wording and full answers. | Question | Guidance shared by the website and API answers | How the presentation varied | | ---------------------------------------------- | -------------------------------------------------------------------------------------------------------------------------------- | ---------------------------------------------------------------------------------------------------------------------------------- | | Monitoring a SaaS brand in AI answers (P02) | Build a set of buyer questions, check them across AI platforms, and track mentions, sentiment, and competitors over time. | The website outlined a practical setup; the API added tool shortlists and more specific tracking steps. | | Choosing an agency reporting tool (P05) | Prioritize coverage across AI platforms, competitor comparisons, scheduled reports, and branded client dashboards. | The website offered a longer evaluation checklist; the API grouped similar criteria around tracking, automation, and agency needs. | | Setting up tracking for multiple clients (P06) | Create prompt libraries for each client, establish a baseline, schedule checks, and connect client reports to business outcomes. | The answers organized the steps differently and varied in suggested tools, prompt counts, and schedules. | | Checking whether AI recommends a product (P07) | Test realistic buyer questions across platforms, record mentions and sources, and repeat the checks over time. | The website gave more attention to product data and discovery; the API gave more detail on manual checks and monitoring tools. | Tool recommendations also had some common ground. For the agency-platform question (P03), the latest website answer and both API answers named Profound, Semrush, and Peec AI. Their first-mentioned tools and the rest of their lists varied. For the client-reporting question (P08), all three answers included Otterly and Peec AI. The website answer opened with Semrush and Scrunch; both API answers opened with Indexly and Profound. Similar overall advice can therefore sit alongside differences that matter to an individual brand's visibility. ## Why details can vary AI answers have natural run-to-run variation: the same question can lead to different wording, examples, and search choices. Our repeated API requests showed this directly, with the question and model held the same. This pattern is consistent with the randomness involved in AI generation and variation in live search. Collection dates add another influence. Over 77 days, available pages, product information, and search results can change. One question retained “2024” in its wording, while a September API answer explicitly discussed sources from 2025–2026. Keeping prompts' time references current helps make comparisons clearer. The sample supports normal answer variation and collection timing as plausible contributors. Their relative contributions remain unmeasured, alongside the possible effects of different models and search settings. ### The numbers in context | Measure | Result in this sample | | --------------------------------------------------------------- | ----------------------------------------------------------------------- | | Average answer length | 4,490 characters for the latest website answers; 3,009 for API answers. | | Exact tool-brand overlap, website versus API | 7.5% on average among the 44 tool brands counted. | | Exact source-page overlap, website versus API | 3.1% on average. | | Exact source-page overlap between repeated API answers | 9.5% on average. | | Exact source-page overlap between the two saved website answers | 75.6% on average. | **Overlap** is the names or pages shared by two answers as a percentage of all distinct names or pages in either answer, averaged across comparisons. These figures describe how often specific names and links recur. The examples above show how shared advice can draw on different names and pages. The brand count covers a fixed list of 44 tools. Several website answers gave general advice without naming tools on that list, which lowers the overall brand overlap. Pairs where both answers named none of the listed tools were excluded from that average. The website repetitions were collected on separate days; the API repetitions were collected in one session. ## Using the results in OpenLens API answers contribute to your brand-mention, citation, and visibility metrics. The **API** label helps you see how each answer was collected. When reviewing a change, consider the answer's meaning alongside the brands and sources it includes. Collection date, answer source, and patterns across several questions and runs all help explain the result. A recurring change across those observations gives you more context for investigating your brand's visibility. Credit usage stays at **one credit per completed Perplexity answer for each prompt and country**. Learn more about the API in [Perplexity’s Agent API documentation](https://docs.perplexity.ai/docs/agent-api/quickstart). ## About the study The sample contains 32 answers: two saved website answers and two new API answers for each of eight questions about AI visibility tools and agency workflows. The API used DeepSeek V4 Flash with web search enabled. The main website-to-API comparison uses the more recent website answer for each question. We selected eight recent questions about AI visibility and agency workflows, excluding questions with personal-business or location context. The findings reflect this small sample and topic area. The website's original model is unknown, and its references were collected differently from the API's citations. Factual accuracy and source quality were outside the study's scope. ## Explore the study data [Download the study data](/assets/perplexity-web-agent-comparison.json) for the exact questions, all 32 answers, source links, collection dates, and the comparison method. P01–P08 identify the questions; W1 and W2 are the newer and older website answers, and A1 and A2 are the two API answers.