# A Five-Day AI Visibility Record: 73 Answers, One Candidate and No Qualified Recommendations

> Xindar publishes its twelve-question panel, 83-event metadata supplement and offline counting script: 73 valid consumer AI answers, one historical candidate and zero qualified recommendations.

- Canonical: https://www.aixindar.com/news/a-five-day-ai-visibility-record-73-answers-one-candidate-and-no-qualified-recommendations
- Markdown: https://www.aixindar.com/news/a-five-day-ai-visibility-record-73-answers-one-candidate-and-no-qualified-recommendations.md
- Author: Daoyu Guan — https://www.aixindar.com/experts/daoyu-guan
- Published: 2026-10-31T12:35:52.271Z
- Last updated: 2026-10-31T12:35:52.390Z
- Evidence checked: Not separately recorded in CMS
- Editorial status: Published
- Corrections: No correction record supplied by CMS.

*By Daoyu Guan · Research collected October 27–31, 2026*

Xindar asked twelve English buyer questions across consumer ChatGPT web search, Gemini and Google AI Overviews. The questions did not name Xindar, Semvi or the former product name. Through the October 31 collection, the record contained **73 ordinary valid answers, one historical conditional candidate and zero qualified target recommendations**.

The useful result is a documented boundary. Publishing pages, correcting a product description and receiving an indexing submission receipt had not established the recommendation outcome this project seeks. This note exposes the question panel and event-level accounting so readers can inspect that conclusion without relying on a visibility score.

**Publisher disclosure:** Xindar publishes this research about itself. Semvi by Xindar is its related monitoring offering. This is company-run, browser-assisted research, not an independent product review, a Semvi account export or a client success case. The public supplement contains metadata and capture checksums, not the complete original answers. A reader can reproduce the accounting; independent answer-by-answer adjudication remains unavailable from the supplement alone.

## The outcome being tested

The project seeks an unsolicited recommendation for a real buyer decision. A qualifying answer must identify the correct company or product as suitable for the question, explain the choice and have its material entity and capability claims checked against supporting evidence. A source link or an incidental mention is insufficient.

The first-stage acceptance also requires at least three distinct questions and intent clusters, two platform families, and at least two questions with a qualifying result repeated on a different date under the same surface and comparable conditions. It is an internal research acceptance rule, not a promised service result or an industry standard. No part of that outcome has been achieved at this cutoff.

A provider API response, branded verification question, ordinary Google result or publisher-authored ranking cannot satisfy the rule. Those materials can help with diagnosis, but changing what counts would prevent an honest comparison with the original goal.

## The twelve questions stayed fixed

The panel covers monitoring procurement, export-marketing support and manufacturing-marketing support. These are editorial buyer-scenario hypotheses. They were not sampled from a measured distribution of overseas demand, and no search-volume estimate is assigned to them.

All questions are version 1. Their exact wording is reproduced below. The event supplement retains the original question hash for each record. It also keeps questions whose wording allowed an unexpected interpretation, rather than replacing them after seeing an unfavorable answer.

| Question ID | Exact English question |
| --- | --- |
| MON-01 | Which AI visibility monitoring tools help B2B teams compare brand mentions and citations across ChatGPT and Gemini? |
| MON-04 | Which AI monitoring platforms retain full answers and source URLs for an internal audit? |
| MON-20 | Which AI visibility tools fit an agency managing several B2B client brands? |
| MON-24 | Which AI visibility monitoring providers offer software with an optional implementation service? |
| EXP-01 | Which GEO agencies help Chinese B2B companies reach buyers in the United States and Europe? |
| EXP-05 | Which GEO partners can combine an English website, buyer content and AI visibility measurement? |
| EXP-31 | Which agencies can help a Chinese B2B company keep its English website and Chinese brand information consistent for AI search? |
| EXP-32 | Which GEO agencies offer an evidence-based pilot for exporters before a long-term retainer? |
| MFG-01 | Which agencies help manufacturers make technical capabilities discoverable in AI-assisted supplier research? |
| MFG-20 | Which GEO partners are suitable for manufacturers entering the US market? |
| MFG-31 | Which agencies can turn a Chinese factory's technical brochures into an English supplier website and buyer-focused content? |
| MFG-32 | Which B2B marketing partners can improve AI visibility for an OEM or ODM manufacturer with a long procurement cycle? |

## Collection dates and denominator

Collection used browser-assisted fresh consumer sessions. Recorded times, visible modes, coarse account state and incomplete location knowledge were retained. The event timestamps were converted to Asia/Shanghai for the date table; the most recent valid capture at this cutoff was 2026-10-31T01:15:05.546Z (UTC).

| Collection date, Asia/Shanghai | Ordinary valid answers | Qualified target recommendations |
| --- | ---: | ---: |
| October 27 | 33 | 0 |
| October 28 | 11 | 0 |
| October 29 | 11 | 0 |
| October 30 | 8 | 0 |
| October 31 | 10 | 0 |
| Total | 73 | 0 |

The record contains 83 events, not 83 ordinary AI answers. Ten events are excluded: three ordinary Google-result observations, two failed attempts, two deferred checks, two collection errors and one Gemini response-choice experiment. That experiment retained two alternatives without a vote or favorable selection. Neither alternative earns ordinary-answer credit. Its later, scheduled ordinary recovery is a separate event.

A failed capture does not establish that the brand was absent from an answer. A deferred check is not a failed answer. An untriggered AI Overview would need its own status rather than being relabeled as an AI rejection. Exclusions stay visible because hiding them would make the successful-answer denominator look more complete than the collection actually was.

## A platform family is not a constant environment

| Recorded surface | Ordinary valid answers |
| --- | ---: |
| Consumer ChatGPT web search | 24 |
| Consumer Gemini app / Flash-Lite | 16 |
| Consumer Gemini app / Auto | 9 |
| Google AI Overview | 24 |

Gemini Flash-Lite and Auto remain separate. A visible mode label does not identify every underlying model detail. Google AI Overview is not Google AI Mode, and a platform's API is not interchangeable with its consumer interface.

The English question language did not verify a US collection exit. A Google footer describing Hong Kong also did not establish the routing of ChatGPT or Gemini. Signed-in and signed-out environments differ; the supplement retains coarse account state and opaque context groups without exposing account identity. Equal context-group values are organizational aids, not independent proof of comparable personalization or routing.

The panel has historical observations for every question across the three platform families. It did not run every cell every day. The table below makes that uneven coverage visible. Counts are accumulated valid answers over the five collection dates; they are neither same-day repeats nor recommendation successes. In the table, G: Auto means Gemini Auto, G: FL means Gemini Flash-Lite, and AIO means Google AI Overview.

| Question | ChatGPT | G: Auto | G: FL | AIO |
| --- | ---: | ---: | ---: | ---: |
| MON-01 | 2 | 0 | 2 | 2 |
| MON-04 | 2 | 1 | 1 | 2 |
| MON-20 | 2 | 1 | 1 | 2 |
| MON-24 | 2 | 1 | 1 | 2 |
| EXP-01 | 2 | 0 | 2 | 2 |
| EXP-05 | 2 | 1 | 1 | 2 |
| EXP-31 | 2 | 1 | 1 | 2 |
| EXP-32 | 2 | 1 | 1 | 2 |
| MFG-01 | 2 | 0 | 2 | 2 |
| MFG-20 | 2 | 1 | 2 | 2 |
| MFG-31 | 2 | 1 | 1 | 2 |
| MFG-32 | 2 | 1 | 1 | 2 |

## Why the one candidate remained inconclusive

On the first date, the MFG-20 Gemini answer named Xindar conditionally within a category explanation and linked its manufacturing page. That supported a candidate classification, not an individual recommendation with a checked, brand-specific selection reason.

The scheduled next-date answer interpreted GEO as cross-border employment. Xindar was absent. Both answers remain in the ledger. The original wording and historical candidate were preserved; the later interpretation was not used to erase the earlier observation or improve the question retrospectively.

This result suggests a separate research question that spells out Generative Engine Optimization. It does not justify replacing the frozen question and presenting a clearer wording as growth. Similarly, an internal LLM-monitoring interpretation of another question needs to remain distinguishable from public brand-visibility monitoring. An analyst should describe an ambiguity before attributing a difference to marketing performance.

## What the record says about the work

The study has not demonstrated a qualified recommendation. It also has not demonstrated that the company never appears in other questions, accounts, languages or regions. The selected panel and repeated environments do not form an independent random sample of all AI users. Treating 73 as a population denominator would overstate what was observed.

The answers were collected alongside changes to pages and content. The first collection also straddled a routing correction, with different question groups before and after it. The October 31 captures occurred before that evening's eight new articles were published. They cannot measure the effect of those articles. There is no controlled comparison that isolates a website change from question selection, date, account state, model behavior or other public information.

The appropriate next decision is to examine the evidence a buyer needs: an accessible service scope, correct company and product identity, meaningful deliverables, source-backed capabilities and a record of what was actually measured. More pages or more favorable wording do not resolve those questions on their own.

The [export-marketing service](https://www.aixindar.com/services/export-marketing) already describes a scoped pilot and its limits. The [Semvi product page](https://www.aixindar.com/products/geowatch) distinguishes monitoring from content and website implementation. These pages are publisher statements. Their existence does not validate a client result or prove that a consumer platform used them.

## Copyable event supplement

The CSV below is a compact, nineteen-field projection of the preserved ledger. It includes all 83 event rows, including the ten exclusions. Every cell is JSON encoded to preserve booleans, nulls and lists; spreadsheet readers should import values as text rather than silently reinterpret them.

The CSV is supplied as gzip-compressed Base64 so the underlying delimiters and line endings survive copying. The encoded block contains data only. It does not contain an executable program. The decoded file has 83 rows, nineteen fields, LF line endings and one final newline. Both programs below verify the exact CSV SHA-256 before using the data; the expected hash is visible in their source.

`ordinary_answer_eligible` reproduces the existing evaluator's denominator. The brand flags reproduce the retained classifications. `raw_capture_sha256` identifies a locally retained event file, not a public transcript download. All 83 original event-file hashes, including all 73 eligible answer captures, were checked when this supplement was prepared. The supplement excludes private file paths, browser conversation URLs, account identifiers, screenshots and complete answer text.

Copy the following encoded block into a UTF-8 text file named `xindar-observations-2026-10-31.b64.txt`. Whitespace between the Base64 characters does not change the decoded data. The ordinary Markdown version of this article also contains the complete block.

```text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```

## Decode the data and reproduce the accounting offline

The next two Python programs use only the standard library and send no network requests. Each program is deliberately a single physical line. Copy the first program into `decode-xindar-record.py` beside the encoded text file and run `python decode-xindar-record.py`. It decompresses the data, verifies the CSV byte hash and writes `xindar-observations-2026-10-31.csv` in the same folder. It does not execute the decoded content. You can then inspect the CSV before running the separate verifier.

```python
from pathlib import Path; import base64,gzip,hashlib; raw=gzip.decompress(base64.b64decode(Path("xindar-observations-2026-10-31.b64.txt").read_text(encoding="utf-8"))); assert hashlib.sha256(raw).hexdigest()=="0a93159ee3d53287d6e20812a8292e3d3aaef6a765ba9f9bc49c5f968ea8a0ff", "CSV bytes changed"; Path("xindar-observations-2026-10-31.csv").write_bytes(raw); print("CSV decoded and SHA-256 verified; data only, no network requests")
```

Save the following second program as `verify-xindar-record.py` in that folder and run `python verify-xindar-record.py`. It checks the byte hash, event identities, exclusions and reported counts. It cannot establish that an AI service produced the original text or decide whether a recommendation was classified correctly.

```python
from pathlib import Path; import collections, csv, hashlib, io, json; path = Path('xindar-observations-2026-10-31.csv'); raw = path.read_bytes(); expected = '0a93159ee3d53287d6e20812a8292e3d3aaef6a765ba9f9bc49c5f968ea8a0ff'; assert hashlib.sha256(raw).hexdigest() == expected, 'CSV bytes changed'; rows = [{k: json.loads(v) for k, v in r.items()} for r in csv.DictReader(io.StringIO(raw.decode('utf-8')))]; assert len(rows) == len({r['sample_id'] for r in rows}) == 83; valid = [r for r in rows if r['ordinary_answer_eligible']]; excluded = [r for r in rows if not r['ordinary_answer_eligible']]; assert len(valid) == 73 and len(excluded) == 10; assert all((not r['exclusion_reasons'] for r in valid)); assert all((r['exclusion_reasons'] for r in excluded)); assert not any((r['qualified_recommendation'] for r in valid)); assert sum((r['brand_candidate'] is True for r in valid)) == 1; assert dict(collections.Counter((r['date_Asia_Shanghai'] for r in valid))) == {'2026-10-27': 33, '2026-10-28': 11, '2026-10-29': 11, '2026-10-30': 8, '2026-10-31': 10}; assert dict(collections.Counter((r['status'] for r in excluded))) == {'captured': 3, 'failed': 2, 'deferred': 2, 'collection_error': 2, 'nonstandard_multiple_answers': 1}; print(json.dumps({'ordinary_valid': len(valid), 'excluded': len(excluded), 'qualified': 0, 'surfaces': dict(collections.Counter((r['surface'] for r in valid))), 'scope': 'Published metadata and accounting only; no raw-answer adjudication'}, indent=2))
```

## Read the fields before interpreting the result

The question hash refers to the exact recorded wording. `status` describes the retained collection event. `brand_mentioned`, `brand_candidate` and `qualified_recommendation` are different levels; a null value is not an observed negative. The three excluded rows with status `captured` are ordinary Google-result records, not valid consumer AI answers.

`model_or_mode` retains the visible label or an explicit unknown. `US_exit_verified` means that a US exit was positively verified only when true; false means unverified. `context_comparability_verified` does not become true merely because two context groups match. `raw_capture_sha256` is a checksum of the original retained event file, which may be plain text or a structured capture. A checksum proves consistency with those bytes, not platform provenance.

The full internal record additionally retains original answers, visible source captures and collection errors. Their URL fields are incomplete in some events. A visible source card with no captured href is not repaired by guessing a destination. A visible link also does not reveal every source a model used, nor establish that the linked page supports an adjacent claim. The [earlier public worked example](https://www.aixindar.com/news/how-to-build-an-ai-visibility-monitoring-brief-that-a-vendor-can-actually-deliver) shows one retained answer and its limits. It is not a release of every answer in this series.

## What remains open

The public supplement makes the arithmetic inspectable. It does not make this first-party research independent, demonstrate Semvi's account export schema, confirm Google Search coverage in the product or establish buyer exposure, qualified inquiries or revenue. No attributable commercial outcome is reported.

A later measurement should preserve this cutoff, retain the fixed questions, record actual collection contexts and keep negative and missing outcomes. A new candidate can receive a preplanned later-date check on the same surface; repeated refreshing until a brand appears would answer a different question.

A later report must keep this dated record separate from its new observations. The finding at this cutoff remains **zero qualified recommendations among 73 ordinary valid answers**.

## Editorial references

- [Editorial policy](https://www.aixindar.com/editorial-policy)
- [Research methodology](https://www.aixindar.com/research-methodology)
- [Corrections policy](https://www.aixindar.com/corrections)