Waikay published a case study showing that a single structured knowledge graph file, installed on its website on April 25, 2026, raised its AI visibility score on one topic by 26 points within 48 hours - outperforming the site's own About page in citation counts on both Gemini and Perplexity's Sonar model.

What an entitymap actually is

The concept is straightforward, though the implications for marketing teams are not. An entitymap is a structured file - two files in practice, an entitymap.html and an entitymap.json - that describes a brand as a knowledge graph. It declares the brand's entities, the products it offers, the people behind them, and the relationships connecting those elements. Each claim in the file is sourced to a specific URL on the site.

That distinguishes it from two formats the SEO industry already knows. A sitemap lists which pages a site contains. An entitymap, according to Waikay, describes what the brand actually means, what it does, and how its parts relate to each other. Schema.org markup annotates individual pages with structured data; an entitymap represents the whole brand as a connected graph, with provenance attached to every claim.

The practical target is live retrieval. Models like Gemini, when answering a query about a brand, do not always rely solely on training data. They fetch pages from the web at query time and synthesize an answer from what they find. That fetch-and-answer cycle means a change that a live retrieval model can read today can affect responses within 48 hours - far faster than traditional signals like backlinks or indexed content, which accumulate over months. The entitymap is designed precisely for that window.

The experiment

Waikay, an AI visibility monitoring platform operated by Dixon Jones, CEO of inLinks and Waikay, deployed the entitymap on waikay.io on April 25, 2026. The deployment was the only change made to the site during the measurement window. Five topics were tracked, each with AI Visibility Score readings going back to early 2026.

The AI Visibility Score, as defined by Waikay, measures the overlap between what an AI says about a brand and what the brand's website actually says. The scale runs from 0 to 100. A score of 95 indicates close alignment between the AI's output and the brand's own content. A score of 58 indicates the AI is drawing heavily on other sources - competitor pages, review sites, outdated coverage, or in some cases nothing identifiable at all.

Scores are measured by querying Gemini in live retrieval mode with prompts structured as: "What do you know about Waikay in relation to [topic]?" The live retrieval mode is the critical variable. It means Gemini fetches pages as the query runs, rather than relying on what was absorbed during model training.

Five topics, five different outcomes

The results across the five tracked topics varied considerably, and that variation is part of what makes the data credible rather than promotional.

Brand overview was already scoring between 91 and 96 before the entitymap went live. It oscillated between 94 and 97 afterward - a gain of approximately 1 point. No meaningful change occurred. That is the expected result for a topic already near the performance ceiling, and according to Waikay, it is what gives the other results credibility. A tool that moved every score uniformly would be less convincing than one that moves only the scores with room to move.

AI brand visibility was rising before the install, climbing from 85 to 92 in the weeks prior. It continued upward, reaching 97. The gain of 5 points cannot be attributed solely to the entitymap, because the pre-install trend was already positive. The entitymap may have contributed; it cannot be isolated as the sole cause.

AI search optimisation showed a different pattern - three readings in the low 80s, drifting slightly downward. The pre-install trajectory predicted a next reading near 80. It landed at 92: a 10-point gain in the first measurement window after the entitymap went live, with no pre-existing upward momentum to explain it.

AI Fact Tracker had already jumped from 48 to 71 in the month before the install - a 23-point rise in 30 days, attributed by Waikay to content and linking changes around that product page. It then added another 10 points in 48 hours. The rate accelerated: 23 points in 30 days before install, 10 points in 2 days after. Waikay notes this suggests the entitymap contributed but cannot call it the sole cause.

AI hallucinations is the most striking topic in the dataset. The score had been falling by roughly 10 points per month for three consecutive months, dropping from 81 to 58 with no identified cause. Gemini was, during that period, increasingly answering queries about "Waikay and hallucinations" using weaker sources. Forty-eight hours after the entitymap went live, the score moved from 58 to 84 - a 26-point reversal. According to Waikay, the entire three-month decline reversed, not partially, but completely.

The logic Waikay applies to explain the speed is specific: new content takes weeks to be indexed; backlinks accumulate over months; neither mechanism operates on a 48-hour timescale. A reversal of that speed points at live retrieval as the mechanism, and live retrieval is precisely what the entitymap is designed to influence.

Citation data: one file beats the About page

Beyond scores, Waikay tracked which specific URLs Gemini and Sonar cited when answering brand queries about waikay.io.

Across 168 total citations in Gemini, several pages clustered near the top - the aio-guide, features page, and blog each held 8.3% citation share. The entitymap.html file held 6.5%. The About page, which conventional SEO playbooks treat as the authoritative source for brand queries, held 3.0%. That makes the entitymap cited 2.2 times more often than the About page on Gemini.

On Sonar - Perplexity's search model - the pattern held across 76 total citations. The entitymap.html registered 3.9% citation share while the About page registered 1.3%, a ratio of 3.0x. The dilution dynamic on Sonar is relevant here: Sonar cited 46 distinct URLs across those 76 citations, compared to a more concentrated pattern on Gemini. That means all individual citation shares are lower on Sonar, but the entitymap held its relative position while the About page suffered the same dilution effect.

The 30KB entitymap file, in both models, outperformed a page that has been the cornerstone of brand SEO strategy for years.

Where the experiment did not work

The entitymap was not submitted to Bing Webmaster Tools before launch, was not added to sitemap.xml, and was not linked from the homepage. According to Waikay, those three omissions - amounting to roughly 20 minutes of work - meant Bing never indexed the file. As a result, ChatGPT, Microsoft Copilot, and Claude never cited it. Those three models rely on Bing-derived signals for their grounding, and a file Bing has not crawled is a file those models have never seen.

This is a significant limitation in the current dataset. The strongest evidence comes from Gemini and Sonar, both of which operate their own crawlers. Evidence from the Bing-dependent stack is entirely absent. Waikay states it plans to correct the submission gap and publish updated cross-model data once Bing has indexed the file.

The case study also flags three things the data does not support. First, on topics that were already rising before install - AI Brand Visibility and AI Fact Tracker - the entitymap's contribution cannot be separated from work already underway. Second, the absence of results on ChatGPT and Copilot reflects an indexation failure, not a format rejection; what those models would do with a properly indexed entitymap remains untested. Third, one result on one surface is not a settled new standard. Wider conclusions depend on replication by practitioners outside Waikay.

Why this matters for marketing and search professionals

The advertising and marketing industry has been watching AI visibility measurement evolve rapidly. HubSpot launched its AEO tool on April 14, 2026, citing a 27% year-over-year decline in organic traffic for its customers. Research analyzed by PPC Land found that 23 factors influence AI citation probability, with brand and entity trust scoring 6.8 and expected to rise. Separate analysis showed that Google exchanges 56% of its AI Mode citation sources weekly, while ChatGPT exchanges 74% - instability that makes any single week's citation data an unreliable basis for strategy.

PPC Land has documented that AI-mediated search increasingly concentrates citation probability in established, high-authority sources, and that the discipline required to cover the full surface area across ChatGPT, Perplexity, Claude, Copilot, and Gemini has grown beyond what any single optimization approach describes. Research on blocking AI crawlers found that even sites that block major AI bots still appear in 70-92% of AI citations, suggesting citation selection is driven by indexation and retrieval quality rather than access controls alone.

The entitymap experiment sits inside this context. It does not address paid visibility in AI systems - a question PPC Land has tracked as a structural challenge for marketing teams uncertain whether AI visibility falls to PPC or SEO functions. What it addresses is whether a brand can give live retrieval models a single, authoritative, structured source to pull from - and whether doing so measurably shifts what those models say.

The 48-hour reversal on the hallucinations topic is the data point that practitioners in the comments of Dixon Jones' LinkedIn post found most compelling. Artur Ferreira, founder of The GEO Lab, noted that the 48-hour reversal is "the part I would replicate first," and added that citation sets for proprietary-concept queries on Perplexity ran "effectively deterministic across a 14-day window, so a signal that moves a sticky set that fast carries weight." Several practitioners commented that they had implemented the format on their own sites shortly after the case study was posted.

Controlled tests published in April 2026 also showed that Google's AI Mode uses a separate content store - described as FastSearch - rather than the live web index used by classic Google Search. That distinction matters here: the entitymap results on Gemini's standalone model are consistent with live retrieval behavior, while AI Mode may operate differently. The case study does not distinguish between Gemini surfaces.

Technical structure and replication

The format itself is open. Waikay published prompt templates on GitHub that allow any practitioner to generate the two required files - entitymap.html and entitymap.json - using any major AI model, filled with their own brand details.

The methodology Waikay recommends is specific about sequencing. Baseline measurements on 5 to 10 topics should be in place for at least 2 to 3 months before deployment. Without a baseline, changes after install are uninterpretable - there is no reference point against which to measure movement. The entitymap should be deployed on a single date with no other content or linking changes made during the measurement window. Bing Webmaster Tools submission, inclusion in sitemap.xml, and a link from the homepage should be done before launch - not after, as Waikay did. The lowest-scoring topics are where observable effects appear fastest; topics near the ceiling will not move, and that is expected behavior, not failure.

According to Waikay, an entitymap should be cited more often than a site's About page for brand queries within 6 to 8 weeks of proper deployment. If it is not, the case study recommends checking indexation status and internal linking before drawing conclusions about the format's effectiveness.

The file size of the entitymap deployed on waikay.io was 30KB - a single structured document. The About page it outperformed in citation counts on both Gemini and Sonar is presumably larger, more frequently updated, and far more prominent in the site's link architecture. The inversion of that hierarchy, in two independent AI models, across a dataset of 168 and 76 citations respectively, is the result practitioners will want to replicate.

Waikay states it will publish follow-up data once Bing indexation is in place. Until that data exists, the strongest claims apply only to Google-indexed AI surfaces.

Timeline

Summary

Who: Waikay, an AI visibility monitoring platform co-founded and led by Dixon Jones, CEO of inLinks. The case study was conducted internally on Waikay's own domain, waikay.io.

What: Waikay deployed a structured knowledge graph file - called an entitymap - on its website and tracked changes in AI Visibility Scores across five topics, as well as citation rates from Gemini and Sonar. The entitymap, a 30KB structured file, was cited 2.2 times more often than the site's About page on Gemini and 3.0 times more often on Sonar. The most significant score movement was a 26-point gain on the AI hallucinations topic within 48 hours of deployment.

When: The entitymap was installed on April 25, 2026. Score data was tracked through June 1, 2026, covering a five-week window. The case study was published on June 4, 2026.

Where: The experiment ran on waikay.io, a publicly accessible website. Results on Gemini and Sonar are confirmed. Results on ChatGPT, Copilot, and Claude are absent, because Bing had not indexed the entitymap file at the time of publication.

Why: As AI models increasingly answer brand queries by fetching and synthesizing live web content, the accuracy and completeness of what those models say about a brand depends on what structured, authoritative content they can retrieve. The entitymap experiment tested whether a single structured file designed specifically for live retrieval could shift AI output measurably - and, on the surfaces where it was indexed, it did.