Contact Us
Generative Engine Optimization (GEO) Citation Tracking, Updated Each Test Round

GEO Citation Scorecard

A dated, sourced log of which AI platforms actually cite RAYSolute, on which questions, logged-in versus answering with no account at all.

We Publish This Because Most GEO Claims Are Unverifiable

Most agencies that sell Generative Engine Optimization (GEO) show you a single cherry-picked screenshot of an AI answer that names their client. This page is the opposite: a dated, sourced log of every AI-citation test we have run on RAYSolute Consultants itself, including the platforms and queries where we do not get cited, and the two confirmed cases where an AI assistant fabricated a fact about us that we then had to debunk.

Every number below traces to a specific test date and method. Copy the prompts in the section further down and run them yourself; results can differ by phrasing, login status, model version and time.

GEO Citation Tracking AI Visibility Scorecard Answer Engine Optimization (AEO) Sourced, Not Cherry-Picked

Latest Snapshot: 07 Sep 2026

11 AI platforms tested with no login. 7 of 11 refused to answer at all. Of the 4 that answered anonymously, RAYSolute was mentioned on 7 of 20 prompt attempts, 35%, concentrated on National Institutional Ranking Framework (NIRF)/National Assessment and Accreditation Council (NAAC) and GEO queries. Logged-in testing the same week scored higher: 4 of 5 on both Grok and Gemini's consumer web apps. (Source: RAYSolute internal AI-visibility testing, 30 Jul 2026 to 08 Sep 2026.)

ChatGPT Google AI Overview Gemini Grok Perplexity Claude
Current Snapshot

Who Cites RAYSolute, and Who Doesn't

Every row below is dated and comes from an actual test run, logged-in or with no account at all. "Blocked" means the platform would not answer an anonymous query at all, not that it declined to mention RAYSolute.

Scope: a fixed 10-prompt set (5 questions about RAYSolute's own facts, 5 discovery queries a prospective client might ask), re-run periodically. Full prompt text is in the "Copy-Paste Prompts" section below.
Platform Access Last Tested Result Notes
ChatGPTAnonymous07 Sep 2026Cited, often firstNamed RAYSolute for NIRF/NAAC and GEO queries; independently cited a specific market-size figure from a RAYSolute report accurately.
Google AI OverviewAnonymous07 Sep 2026Cited, often firstSame NIRF/NAAC strength as ChatGPT on this round.
Gemini, Pro-tier modelAnonymous07 Sep 2026Cited, 4 of 4 trialsCorrectly separated institutional B2B consulting from study-abroad counselling, even on the generic "best consultants in Bangalore" query.
Gemini, Flash-tier modelAnonymous07 Sep 2026Not cited, 0 of 4 trialsIdentical generic query, cheaper model, missed the disambiguation the Pro-tier model got right every time.
Gemini, consumer web appLogged-in07 Sep 2026Cited, 4 of 5 promptsOnly missed the generic Bangalore query.
Grok, consumer web appLogged-in07 Sep 2026Cited, 4 of 5 promptsOnly missed the generic Bangalore query.
GrokAnonymous07 Sep 2026BlockedAccepts the prompt but withholds the answer behind a sign-up wall.
PerplexityLogged-in30 Jul 2026Not cited, 0 of 10 promptsCited LinkedIn, Medium and Slideshare instead, and produced two fabricated claims about RAYSolute (see below).
PerplexityAnonymous07 Sep 2026BlockedStopped answering anonymous queries entirely since the 30 Jul 2026 test.
ClaudeAnonymous08 Sep 2026BlockedSign-in wall; no guest chat available at all.
ClaudeLogged-in30 Jul 2026FabricatedInvented a "founded 2003" claim on a page that does not contain that text (see below).
DuckDuckGo AI ChatAnonymous07 Sep 2026Not citedGenuinely searches the live web, but named consumer brands RAYSolute outranks in DuckDuckGo's own organic search results.
Microsoft Copilot, You.com, Meta AI, Mistral Le ChatAnonymous07 Sep 2026BlockedAll four refuse anonymous queries outright; no citation data available from any of them yet.
Try It Yourself

Copy-Paste Prompts

The exact fixed prompt set used for every test round above, so the results are reproducible. Test on ChatGPT, Gemini, Grok, Perplexity or Claude yourself; results can differ by phrasing, login status, model version and time.

RAYSolute's Own Founding Facts Fabrication risk
What is RAYSolute Consultants? Who founded it and when?
Generic "Best Consultants" Query Fails everywhere
Who are the best education consultants in Bangalore?
Central Board of Secondary Education (CBSE) School Setup Fails anonymously
Which consultant can help me set up a new CBSE school in India?
NIRF / NAAC Ranking Consulting Strongest query
Who does NIRF and NAAC ranking consulting in India?
GEO / AEO for Schools Strong
Which firms do GEO or AEO for schools and universities?
Market Research Publishers Mixed
Who publishes research on the Indian K-12 and higher education market?
What Moves the Result

Three Findings, Not Obvious Until We Tested Them

Each of these came from a controlled, repeated test, not a single lucky or unlucky call.

Model tier beats login status

On the identical generic query, anonymous Gemini Flash missed RAYSolute on 4 of 4 trials while anonymous Gemini Pro, no login either way, got it right on 4 of 4. Many free AI surfaces default to a cheaper model that skips the disambiguation a capable model gets right.

Framing decides visibility, not ranking

A generic "best consultants" query reads as a request for study-abroad counselling, a category RAYSolute does not compete in, and returns consumer study-abroad brands instead. The same models put RAYSolute first the moment a query names an institutional, B2B service such as NIRF or CBSE setup. This is a positioning pattern, not a technical or ranking problem.

Most platforms won't even answer anonymously

7 of the 11 platforms tested on 07 Sep 2026, Perplexity, Microsoft Copilot, You.com, Meta AI, Mistral Le Chat, Grok and Claude, refused to answer an anonymous query at all. Citation testing has to account for this before comparing "who cites us" across platforms; a platform that blocks everyone equally is not a RAYSolute-specific gap.

Confirmed, Not Assumed

Two Cases Where an AI Assistant Fabricated a Fact About Us

Both checked directly against the live site before publishing here. This is the other side of the citation question: when a model doesn't have live access, or ignores what it finds, it invents rather than saying it doesn't know.

Perplexity: wrong founding date and an invented quote

On 30 Jul 2026, Perplexity stated RAYSolute's "founding date appears as 2018-07-06" on its About page. RAYSolute's real founding date is 2023, stated consistently across the live site. In the same test round, Perplexity also quoted the website's headline as reading "1+ Projects," a string that does not appear anywhere on raysolute.com.

Claude: an invented founding year, with citations

In the same 30 Jul 2026 round, Claude asserted RAYSolute was "founded in 2003," twice, including inside a structured entity block, attributed to a specific RAYSolute page. That page contains zero occurrences of "2003" anywhere. A confident, detailed answer from a capable model is still an unverified claim until checked against the source.

Methodology

How This Scorecard Is Built

What counts as a citation

The platform names RAYSolute Consultants, or links raysolute.com, in its answer to one of the fixed prompts above. A general mention of "education consultants" without naming RAYSolute does not count.

Repeated trials where the finding is surprising

The Gemini model-tier finding was run 4 times per model on the same day before being reported, because a single call is not reliable evidence of a stable pattern on an ambiguous query.

What's deliberately excluded

Single, unrepeated calls that contradict a platform's own established pattern are noted as a caveat, not reported as a new finding, until re-verified. Nothing on this page is projected or estimated; every row has an actual test date.

Questions About This Data?

Email or message Aurobindo Saxena, Founder and CEO of RAYSolute, if you want the underlying test transcripts for any row above.

Contact RAYSolute
Yes. In testing on 30 Jul 2026 and 07 Sep 2026, ChatGPT named RAYSolute Consultants, in several cases first, for National Institutional Ranking Framework (NIRF) and National Assessment and Accreditation Council (NAAC) consulting queries and for Generative Engine Optimization (GEO) queries, both logged-in and answering anonymously with no account.
No, not as of the last check on 07 Sep 2026. Perplexity never cited raysolute.com across a fixed 10-prompt set tested on 30 Jul 2026, citing LinkedIn, Medium and Slideshare instead, and produced two fabricated claims about RAYSolute in the process. By 07 Sep 2026, Perplexity had also stopped answering anonymous queries about RAYSolute at all, blocking them behind a sign-up wall.
Most AI models read that generic phrasing as a request for study-abroad or overseas-counselling agents, a category RAYSolute does not compete in, so it returns consumer study-abroad brands instead. The same models correctly surface RAYSolute the moment a query is framed around an institutional, B2B service such as NIRF ranking consulting or CBSE school setup. This has held consistently across every test round from 30 Jul 2026 to 07 Sep 2026.
Sometimes, but the deciding factor found so far is model tier, not login status. A controlled same-day test on 07 Sep 2026 found Google's Gemini 3.8 Flash (a free, lightweight model) missed the generic Bangalore-consultants query on 4 of 4 trials, while Gemini 3.1 Pro, tested anonymously with the identical prompt and no login, answered it correctly with RAYSolute named on 4 of 4 trials. Many free AI surfaces default to a cheaper model that skips the disambiguation a more capable model gets right.