Case study, April 2026

We Installed an Entitymap on Waikay.io. Here’s What Happened.

TL;DR

Waikay.io installed an entitymap, a structured machine-readable brand knowledge graph file, on April 25, 2026, and measured its impact on how AI models answer queries about the brand. By June 1, across Google-indexed AI surfaces (Gemini and Sonar/Perplexity), results were strong: AI Knowledge Scores improved by up to 26 points in as little as 48 hours, and the entitymap was cited 2.2-3.0x more often than the site’s own About page. Results on Bing-dependent models (ChatGPT, Copilot) are still pending indexation.

+26 pts
AI Visibility Score lift on hallucinations topic in 48 hours
2.2x
Entitymap cited more than About page (Gemini)
3.0x
Entitymap cited more than About page (Sonar)
💡
Section 01

What Is an Entitymap and Why Does It Exist?

When an AI answers a question about your brand, it does not read your website the way a person would. It looks for entities: specific facts, relationships, and claims it can retrieve and stitch together into an answer. If those entities are scattered across unstructured pages, the AI fills in the gaps from wherever it can: competitor content, review sites, old press coverage, or nothing at all.

An entitymap is a single structured file that describes your brand as a knowledge graph. It lists your entities (the brand, its products, key concepts, the people behind it), the relationships between them (PRODUCED_BY, OFFERS, TARGETS, MEASURES), and source-tagged content chunks that tie each fact to a specific URL on your site.

📄

Not a sitemap

A sitemap tells crawlers which pages exist. An entitymap tells AI models what your brand means, what it does, and how its parts relate to each other.

🔗

Not Schema.org

Schema.org marks up individual pages. An entitymap describes the whole brand as a connected graph, with every claim sourced to a URL.

🤖

Built for live retrieval

Models like Gemini fetch pages at query time. An entitymap gives them one authoritative source rather than fragments from across the web.

What we wanted to find out

We deployed an entitymap on waikay.io on April 25, 2026 and tracked what happened to our AI Visibility Scores across five topics over the following five weeks. This case study is the result. The entitymap was the only change we made during the measurement window.

🗺️
Section 02

What Is an AI Visibility Score?

When an AI like Gemini or ChatGPT answers a question about your brand, it draws on entities: specific facts, concepts, products, and claims it has retrieved about you. Your AI Visibility Score measures how much of what the AI says about you matches what is actually on your website.

Your website What AI says Score

Think of it as a Venn diagram. The left circle is everything on your website. The right circle is everything the AI says about you. The score (0-100) measures the size of the overlap. A score of 95 means the AI’s picture of your brand closely matches your own. A score of 58 means a lot of what it says is missing, wrong, or pulled from somewhere else.

How we measure it

We query Gemini in live retrieval mode with prompts like “What do you know about Waikay in relation to [topic]?” Live retrieval means Gemini fetches pages from the web at query time rather than relying only on training data. A change a live retrieval model can read today can affect answers within 48 hours, far faster than anything requiring a model to be retrained.

We track scores monthly on five topics, with readings going back to early 2026. The entitymap went live on April 25, 2026 as the only change we made during the measurement window.

📊
Section 03 Key data

Five Topics, Five Different Stories

Each chart shows the AI Visibility Score history for one topic. The Y-axis is zoomed to the relevant range so changes are clearly visible. The red dashed line marks April 25, when the entitymap went live.

Brand overview
Already at ceiling. No change expected, none observed.
~+1 pt
100 95 90 85 80 Install Feb Mar Apr May Jun

The score oscillated between 91 and 96 for three months before install, and between 94 and 97 after. There was no room for meaningful improvement and nothing meaningfully improved. This is the expected result, and it is what gives the other findings credibility.

AI brand visibility
Already rising before install. Trend continued upward.
+5 pts
100 95 90 85 80 Install Feb Mar Apr May Jun

This topic was trending upward before install and kept climbing afterward, reaching a new high of 97. Because the score was already rising, we cannot credit the entitymap with the full lift.

AI search optimisation
Three readings drifting down. Jumped after install.
+10 pts
100 95 90 85 80 Install Feb Mar Apr May

Three readings flat in the low 80s, drifting slightly downward. The prior trend predicted the next reading would land near 80. It landed at 92: a 10-point jump in the first window after install, with no pre-existing momentum to explain it.

AI Fact Tracker
Already climbing fast before install. Rate accelerated after.
+10 pts in 2 days
100 90 80 70 60 50 40 Install Jan Feb Mar Apr

The score had already jumped from 48 to 71 in the month before install, probably from content and linking changes around the Fact Tracker product page. It then added another 10 points in 48 hours. The pre-install rate was +23 in 30 days. The post-install rate was +10 in 2 days, faster per day, suggesting the entitymap contributed. We cannot call it the sole cause.

AI hallucinations
Three months of decline. Fully reversed within 48 hours of install.
+26 pts in 48 hrs
90 85 80 75 70 65 60 55 Install Jan Feb Mar Apr
84 Score

58 to 84 in 48 hours

The score had been falling by roughly 10 points a month for three months with no clear cause. Gemini was increasingly answering “Waikay and hallucinations” queries using weaker sources. Forty-eight hours after the entitymap went live, it reversed the entire decline. New content takes weeks to index. Backlinks accumulate over months. A 48-hour reversal points at live retrieval as the mechanism, and that is exactly what the entitymap is designed to influence.

All five topics at a glance
TopicPre-install trendScore changeSpeed
Brand overviewHigh and stable (91-96)~+1n/a
AI brand visibilityRising (85 to 92)+5Weeks
AI Fact TrackerRising fast (48 to 71)+102 days
AI search optimisationFlat and drifting down (86 to 82)+1010 days
AI hallucinationsCollapsing (81 to 58)+262 days
The pattern

The entitymap pulls scores toward a 90-97 ceiling. Topics already there stay there. Topics far below it, particularly those that were flat or actively declining, see the largest and fastest improvements. The effect is strongest precisely where Gemini had been pulling weaker sources into its answers.

🔗
Section 04

What’s Actually Getting Cited

Beyond scores, we tracked which URLs Gemini and Sonar (Perplexity) cite when answering brand questions about waikay.io. The entitymap row is highlighted in both tables.

Gemini: 168 total citations
PageShare
aio-guide8.3%
features8.3%
waikay-blog8.3%
entitymap.html6.5%
metrics-to-track6.0%
brand-visibility-tracker5.4%
reviews4.8%
academy-learning4.2%
about3.0%
Sonar: 76 total citations
PageShare
features6.6%
technology-arch5.3%
aio-guide5.3%
entitymap.html3.9%
ai-audit3.9%
waikay-blog3.9%
faq3.9%
fact-tracker2.6%
about1.3%

The About page is the URL every SEO playbook says should win brand queries. In both models, a single 30KB structured file is cited more often: 2.2x on Gemini, 3.0x on Sonar.

Why the Sonar ratio is larger

Sonar cites more pages per answer than Gemini: 46 distinct URLs across 76 citations, versus Gemini’s more concentrated pattern. That pushes all individual citation shares down on Sonar. The entitymap’s 3.9% looks lower than its 6.5% on Gemini, but the ratio against the About page is actually larger, because the About page suffers the same dilution while the entitymap holds its position.

🚧
Section 05

Where It Doesn’t Work Yet

The entitymap is not indexed by Bing. That single fact explains why ChatGPT, Copilot, and Claude never cite it. Those models are not rejecting the format. They have never seen the file. Until Bing crawls it, we have no evidence about how the format performs on those surfaces.

Gemini
Citing
Google index. Top tier citations.
AI Mode
Citing
Google index. Regular pickup.
Sonar
Citing
Own crawler. Beating About page 3x.
ChatGPT
Not citing
Relies on Bing. Not indexed.
Copilot
Not citing
Relies on Bing. Not indexed.
Claude
Not citing
Bing-derived signals. Not indexed.
The fix is straightforward

Submit via Bing Webmaster Tools, add to sitemap.xml, and link from the homepage. We skipped all three before deploying. It is roughly a 20-minute job. We will publish updated cross-model data once it is done.

Three things the current data does not support:

⚠️

Sole cause on rising topics

AI Brand Visibility and AI Fact Tracker were already climbing before install. We cannot separate the entitymap’s contribution from the work already underway.

⚠️

Works on all models

Strong evidence on Gemini, good evidence on Sonar, zero evidence on the Bing stack, because Bing has not indexed the file.

⚠️

A settled new standard

One strong result on one surface is not a standard. Wider conclusions depend on replication by other practitioners.

Section 06

If You Want to Test This Yourself

1

Create your entitymap files

You need two files: entitymap.html and entitymap.json. Use the prompt templates from the entitymap project to generate both with any major AI model.

Prompt template Copy the prompt, fill in your brand details, and paste the output into the appropriate files on your site. View the full prompt template on GitHub →
2

Baseline first

Track AI Visibility Scores on 5-10 topics for at least 2-3 months before you deploy anything. Without a baseline, post-install changes are uninterpretable.

3

Deploy on one date, change nothing else

A clean before/after requires a single intervention. Do not run other content or linking changes in the same window.

4

Submit to Bing before you launch

Bing Webmaster Tools submission, sitemap.xml inclusion, and a homepage link. Do this first. We did not, and the Bing-stack models never saw the file.

5

Watch the low-scoring topics

That is where the effect shows up fastest. Topics already near the ceiling will not move. That is the expected result, not a failure.

6

Your entitymap should beat your About page within 6-8 weeks

If it is not being cited more often than your About page for brand queries by then, check indexation and internal linking first.

Find out if Gemini is citing you, or something else

We are publishing a follow-up once the Bing indexation is in place. The results so far are strong enough to act on. They are not strong enough to call settled. Waikay Entitymap is coming as a paid service. Join the waitlist to be first in line.

Read the entitymap documentation →