Research · CitePulse.ai — by GeoFlow Solutions
Why a technically readable site still gets no AI citations: citepulse.ai measured on itself
Stable address: this page. Data window: 2026-09-05 to 2026-09-09. Last data refresh: 2026-09-09. New measurements are added as dated updates at the bottom; the conclusions never go beyond the questions listed here.
Short answer. Between 2026-09-05 and 2026-09-09 we asked five AI engines the same 6 buying questions about AI-visibility tools, three times per day per engine. Across 351 measured answers, citepulse.ai was named in the answer text or cited as a source 0 times. The engines cited 1628 URLs for these questions; 0 were on citepulse.ai. The site passes every technical readability check we run (robots, schema, llms.txt, crawlability). Readability is not the reason. The pages the engines actually cite are comparison articles, vendor pages and Reddit threads, listed below.
1. The questions, exactly as asked
English, asked verbatim, no brand name in the question. Frozen since 2026-09-05 so every day is like-for-like.
- best AI visibility monitoring tool for small B2B companies
- best tool to check if my company is showing up in AI-generated answers
- how can I see if AI assistants are mentioning my competitors more than me
- how do I audit how often my brand is being cited by AI tools like ChatGPT
- how do I find out what AI chatbots are saying about my business
- is there a way to audit my visibility in large language model responses
2. The engines, the models and how they were accessed
| Engine (label in the product) | Exact model, access and search |
|---|---|
| Perplexity | model sonar via the Perplexity API; the model searches the web on every answer and returns its sources |
| Gemini | model gemini-2.5-flash via the Gemini API with the google_search grounding tool; sources = grounding chunks |
| ChatGPT (GPT-4o family) | model gpt-4o-mini via the OpenAI Responses API with the web_search_preview tool. This is the API model, not the consumer ChatGPT app; the product labels it "ChatGPT" |
| Claude | model claude-sonnet-4-6 via the Anthropic API with the server-side web_search tool (max 3 searches per answer) |
| Google AI Mode | Google AI Mode answers fetched through the DataForSEO SERP API (location United States, language en); no model name is exposed. Added 2026-09-08 |
Repetitions: 3 per question per engine per day (the daily instrument panel), plus the weekly monitoring run for citepulse.ai, which asks the same questions and adds one more. Locale: English; DataForSEO calls use United States / en. Each answer is scored by the published detection rule: (a) brand name in the answer text, (b) citepulse.ai in the answer's sources. Rows in results.csv carry the verbatim excerpt whenever (a) fired.
3. Results per day and engine
Cells show answers naming or citing citepulse.ai / answers measured. A dash means the engine returned no answers that day (see gaps).
| Date (UTC) | Perplexity | Gemini | ChatGPT (GPT-4o family) | Claude | Google AI Mode |
|---|---|---|---|---|---|
| 2026-09-05 | 0 / 21 | — | 0 / 30 | 0 / 15 | — |
| 2026-09-06 | 0 / 21 | — | 0 / 30 | 0 / 15 | — |
| 2026-09-07 | 0 / 15 | — | 0 / 15 | 0 / 14 | — |
| 2026-09-08 | 0 / 33 | 0 / 14 | 0 / 33 | 0 / 15 | 0 / 6 |
| 2026-09-09 | 0 / 15 | 0 / 15 | 0 / 15 | 0 / 14 | 0 / 15 |
Split by rule: mentions in answer text (a): 0; domain citations in sources (b): 0; entity check (c): not applicable, nothing to disambiguate. There is no namesake problem in this data because there is no mention at all.
4. What the engines cited instead
Top cited domains across all 1628 cited URLs (full list in sources.csv). "Engines" is how many of the five engines cited the domain at least once; "questions" how many of the 6 questions it appeared under.
| Domain | Citations | Engines | Questions |
|---|---|---|---|
| semrush.com | 39 | 3 | 6 |
| reddit.com | 39 | 3 | 5 |
| otterly.ai | 27 | 3 | 6 |
| llmpulse.ai | 26 | 4 | 5 |
| therankmasters.com | 25 | 3 | 5 |
| trysight.ai | 24 | 3 | 5 |
| dageno.ai | 24 | 3 | 5 |
| seranking.com | 23 | 2 | 5 |
| ahrefs.com | 23 | 3 | 3 |
| siftly.ai | 21 | 3 | 5 |
| useomnia.com | 16 | 2 | 5 |
| meltwater.com | 16 | 3 | 5 |
| youtube.com | 14 | 2 | 6 |
| vismore.ai | 14 | 2 | 4 |
| nightwatch.io | 14 | 2 | 5 |
| integrate.io | 14 | 3 | 5 |
| indexly.ai | 14 | 3 | 3 |
| blog.hubspot.com | 14 | 3 | 4 |
| wellows.com | 13 | 4 | 3 |
| frictionai.co | 13 | 2 | 3 |
| elfsight.com | 13 | 3 | 5 |
| airops.com | 13 | 3 | 3 |
| visible.seranking.com | 12 | 2 | 4 |
| qoulomb.com | 12 | 2 | 4 |
| keyword.com | 12 | 3 | 4 |
What this list is: the pages a buyer of an AI-visibility tool is actually sent to. Vendor pricing pages and comparison articles dominate; Reddit and YouTube appear for "how do I" questions. What it is not: a ranking of tool quality.
5. Gaps and limits
- Gemini returned no answers on 5–7 September (the API credits had run out; restored 8 September). Those days show — for Gemini, not 0.
- Google AI Mode joined the panel on 2026-09-08; earlier days have no AI Mode column.
- The Claude and OpenAI numbers come from API models with web search, not from the consumer apps. Consumer ChatGPT can answer differently; we do not claim otherwise. The API model behind the "ChatGPT" label is gpt-4o-mini.
- The questions are ours, in English, US locale. A different question set or locale gives a different source list.
- Three repetitions per day is enough to see day-to-day movement, not enough to estimate a rate more precisely than roughly ±1 answer in 3.
- Zero here means "not once in 351 answers to these 6 questions". It does not mean no engine ever cites citepulse.ai for any question.
6. What we do with it
The site is the wrong lever for these questions: the sources the engines trust are third-party comparison pages. The work is to be on those pages honestly (independent reviews, comparisons that list us, a public report people can inspect), and to keep measuring. Every change we make is logged in the product's Action Center and read against the engines' own day-to-day noise (NOISE-1.0), so a later increase is not mistaken for random movement.
Companion material: the real Full Audit report of citepulse.ai (no form, no demo data) and the methodology. The daily noise panel and this study have different purposes: the panel measures how much the engines move on their own; this study measures whether one site earns citations.
Update log
- 2026-09-09 — Data refreshed: 5 measurement days, 351 answers, 1628 cited URLs.
Data license: CC BY 4.0. Cite as: GeoFlow Solutions / CitePulse.ai, "citepulse.ai measured on itself", 2026-09-09, https://citepulse.ai/research/citepulse-ai