AI Citation Index October 2026: A Four-Platform Pilot

This AI citation index begins with one fixed-prompt test from September 9, 2026. ChatGPT, Gemini and Perplexity cited anobee.com. Google AI Mode did not.

Across the four products, three returned at least one Anobee URL. The successful responses cited 13 distinct pages after duplicate URL variants were normalized. That makes the pilot result 3 out of 4, or 75%.

But the denominator matters. This AI citation index covers one branded prompt, one day and one Nepal-based test session. It is an auditable observation, not a platform benchmark or a claim that Anobee earns citations 75% of the time.

Research note: The test was completed on September 9, 2026. This article is the first edition, so it defines the baseline prompt and classification rules for later monthly comparisons. The target URL was not live during preparation; its index status was therefore unknown before publication.

AI citation index results at a glance

Every platform received the same instruction in a fresh session:

Search the web for Anobee and cite the sources you use.

However, the URL counts cannot be added to get the total because the platforms shared some sources. After normalization and deduplication, the union contains 13 Anobee pages.

Also, two patterns stand out. First, first-party pages that establish identity and editorial standards were useful across the test. Second, a source-rich answer can still be wrong at the entity level. Perplexity cited seven Anobee pages, yet it also pulled in two similarly named but unrelated sites.

The complete AI citation index dataset

Accordingly, the tables in this article are the complete published dataset for the pilot. A URL counts once per product after removing tracking parameters and treating trailing-slash variants as the same page.

Google AI Mode is absent from the URL table because it returned no Anobee source for the fixed prompt.

What ChatGPT cited

ChatGPT returned the broadest topical view of Anobee. Its six normalized first-party sources included the homepage, About page and SEO hub, followed by three guides about generative engine optimization, AI citation checking and the differences among SEO, GEO and AEO.

Also, ChatGPT cited two external pages: a Nepal SEO article and a LinkedIn post. Those sources are not counted as Anobee URLs, but they matter when reviewing how the answer assembled its description.

Meanwhile, the response initially showed one guide in both trailing-slash and non-trailing-slash forms. This index treats those as one normalized URL. Counting visual citation markers instead would have inflated the page total.

ChatGPT's source mix was the most useful for a reader trying to understand both the publication and its subject coverage. Still, a citation is not an endorsement or a guarantee of accuracy. OpenAI advises users to open and check cited sources because web results can be incomplete, outdated or incorrect [1].

What Gemini cited

Gemini used three first-party pages: the homepage, About page and Contact page. Two appeared as inline citations, while the Contact page appeared in the accompanying source list.

Still, that compact set was enough for the response to identify Anobee correctly. Unlike ChatGPT, Gemini did not surface a topical guide in this run.

However, Gemini's interface requires careful classification. Google explains that links displayed by its double-check feature are not necessarily sources used to generate the original response [2]. For this pilot, only links attached to the response or included in its source display were recorded. Later verification links were not added.

What Perplexity cited

Perplexity cited seven Anobee pages, the highest raw count in the pilot. The mix was unusual: About and Start Here appeared beside the terms, privacy, editorial, cookie and fact-checking policies.

Also, those pages give a system structured statements about ownership, purpose and publishing standards. Their presence suggests that clear entity and trust pages can participate in branded answer construction. This result does not prove that any particular policy page caused the selection.

Additionally, Perplexity returned the Anobee X profile, which is a first-party social property but not an anobee.com page. More importantly, it included anabeeb.com and support.anabei.com. Both are unrelated to Anobee. The answer still described Anobee largely correctly, but the source set contained entity collisions.

Consequently, that is a meaningful failure. A long source list may look authoritative while quietly blending separate organizations. Perplexity describes its product as web search with citations to source material [3]; the user still needs to inspect whether each cited entity belongs in the answer.

Why Google AI Mode recorded a zero in the AI citation index

Google AI Mode did not answer the intended brand question. It interpreted “cite the sources you use” as a request for tools that create academic citations and returned sites such as Scribbr, Citation Machine, MyBib and BibMe.

Accordingly, no Anobee page appeared, and the response did not identify the brand. The zero stays in the dataset.

However, changing the prompt to “What is Anobee?” produced a different result during exploration, but that answer is excluded from the four-platform comparison. Replacing a failed run with friendlier wording would break the fixed-prompt method.

Also, the outcome shows why prompt logs need raw answers, not only citation counts. A zero caused by topic interpretation is different from a response that understands the brand but chooses other sources. Google notes that AI Mode can make mistakes and encourages users to try different query versions when a response misses the intent [4]. That is sound search advice, but a longitudinal benchmark still needs to preserve the original observation.

What the AI citation index tells Anobee

However, this AI citation index does not reveal an AI ranking formula. It does highlight practical work worth keeping.

Identity pages supported branded retrieval

The About page was the only Anobee URL shared by ChatGPT, Gemini and Perplexity. The homepage appeared in two products. Contact, Start Here and policy pages widened the first-party evidence available about the brand.

Therefore, identity maintenance is a sensible priority. Names, ownership, purpose and editorial standards should agree across these pages. Contradictions would give retrieval systems and readers competing versions of the same entity.

Topical guides appeared in only one product

ChatGPT alone cited Anobee's SEO and AI-search guides in the fixed run. That does not mean the other platforms cannot cite them. It means they did not do so for this branded prompt on this date.

Therefore, non-branded discovery needs its own prompt panel. Mixing questions such as “What is Anobee?” with “How do I check AI citations?” would make one headline rate difficult to interpret. Future editions should report the branded baseline and any non-branded topic panel separately.

Entity collisions deserve their own error field

Perplexity's two unrelated domains show why “cited or not cited” is too crude. The monthly log also needs an entity-accuracy field.

Useful labels include:

• correct brand and correct sources
• correct brand with unrelated sources mixed in
• wrong entity
• prompt misinterpreted
• no answer or no citation

This classification keeps a high citation count from hiding a weak response.

AI citation index pilot methodology

The goal was reproducibility, not scale. One exact branded prompt was tested across four public products on the same date.

Test conditions

Still, the products did not offer identical account, personalization or search controls. Those differences are recorded rather than treated as equivalent.

Counting rules

A platform passed the citation test when its original answer displayed at least one source link resolving to anobee.com. A brand mention without such a link would not count.

Each normalized page counted once per product. Tracking parameters were removed, and trailing-slash variants were merged. First-party social profiles were logged but excluded from the anobee.com total. Unrelated lookalike domains were recorded as errors, not Anobee citations.

The failed Google result was not rerun until it passed. Exploratory prompts with different wording were excluded from the headline figure.

What this AI citation index cannot prove

The pilot has a sample of one prompt. It cannot estimate:

• Anobee's visibility for non-branded topics
• each product's overall citation frequency
• how results change across countries, languages or accounts
• whether any page characteristic caused a citation
• whether the same response will repeat tomorrow
• referral traffic or the business value of a citation

The 75% figure is therefore descriptive: three successful products divided by four tested products for this exact run. It should not be compared with citation-rate studies that use hundreds or thousands of prompts.

There is also a timing limit. The October 2026 edition reports a pilot run completed on September 9 so the article can be prepared ahead of its planned publication date. Every table names the actual observation date.

What will stay fixed in the AI citation index next month

The branded baseline prompt will remain unchanged. Later editions will keep the same normalization and error rules, show the actual test date and retain zero results.

If Anobee adds a non-branded prompt panel, it will appear as a separate dataset with its own denominator. That panel can measure discovery around topics such as AI citations, generative engine optimization and SEO for Nepal without contaminating the branded baseline.

Four actions will make the series more useful over time:

1. archive the raw answer and source display for every run
2. report added, retained and lost URLs by product
3. flag name collisions and prompt misunderstandings separately
4. connect citation observations with referral and Search Console data where available

For a practical method you can use on your own site, follow Anobee's guide to checking AI citations. The broader generative engine optimization guide explains how technical eligibility, clear answers and credible evidence fit together.

Subscribe for the next AI citation index to see whether the same branded prompt keeps, gains or loses sources.