GEO Lab #03 · S3-CHATGPT-M1-R1 · Original response

This page presents one selected response from GEO Lab #03. It adds a smaller reading and citation entry point; it does not replace the full archive or collect a new answer.

The OpenAI arm uses the Responses API with web search and does not represent consumer ChatGPT Search. Grok records identify their actual consumer-interface mode and available model label.

Recorded conditions

Record ID
S3-CHATGPT-M1-R1
Platform
OpenAI API + web search
Question / condition
M1
Repeat
1
Model label
gpt-5.6-sol
Visible mode
Not independently recorded
Reasoning
medium
Submitted / first recorded time
2026-11-02T09:24:46.262440+00:00
Completeness
Recorded body complete
Search
Actual search activity observed; exact controls and evidence remain in the raw record.

Original prompt

I'm a solo SaaS founder with no GEO budget. How can I reliably measure whether ChatGPT and Grok recommend my product to potential buyers?

The agreed prompt is reproduced exactly. Any recorded difference in the actual submitted text or search UI is retained in the raw record and full appendix.

Original answer

You can’t observe all private ChatGPT or Grok conversations. What you **can** measure reliably is:

1. **Recommendation visibility** across a fixed sample of realistic buyer prompts.
2. **Consistency** across repeated runs.
3. **Citations and referral traffic** reaching your site.
4. **AI-attributed signups and revenue.**

## 1. Build a buyer-prompt benchmark

Create 12–20 prompts based on questions actual prospects ask. Don’t include your product name unless testing brand knowledge.

Cover these intent types:

| Intent | Example |
|---|---|
| Category discovery | “What are the best `[category]` tools for a 10-person SaaS?” |
| Problem-first | “How can I solve `[problem]` without `[undesired workaround]`?” |
| Alternatives | “What are good alternatives to `[competitor]` for `[use case]`?” |
| Comparison | “Compare the top tools for `[job]`, including pricing and limitations.” |
| Constraint-based | “Recommend a `[category]` tool under `$X/month` that supports `[feature]`.” |
| Persona-specific | “I’m a `[role]` at a `[company type]`. What should I use for `[job]`?” |
| Purchase decision | “Which product would you choose for `[specific scenario]`, and why?” |

Source the wording from:

- Sales and support conversations
- Search Console queries
- Customer interviews
- Competitor review pages
- Questions posted in relevant communities

Weight purchase-intent and ICP-specific prompts more heavily than broad “best tools” queries.

## 2. Use a controlled testing protocol

For every benchmark run:

- Start a **new conversation**.
- Turn off memory/custom instructions where possible.
- Keep the same location, prompt wording, and product constraints.
- Record the date, platform, mode, and visible model.
- Test both ordinary chat and explicit web search when practical.

This matters because ChatGPT can use memory, custom instructions, conversation context, and approximate location when forming results. ChatGPT’s search mode may cite sources, while Grok can search both the web and public X content. ([help.openai.com](https://help.openai.com/en/articles/9237897-searching-the-web-with-chatgpt))

Run each prompt **three times per platform**. A single response is too stochastic to treat as a ranking.

A reasonable no-budget cadence:

- **Weekly pulse:** Five most important prompts, twice each
- **Monthly benchmark:** All prompts, three times each
- **After major launches:** Rerun relevant prompts

## 3. Score every response with a simple rubric

Use a spreadsheet with one row per response:

| Field | What to record |
|---|---|
| Platform | ChatGPT or Grok |
| Mode | Default or web/search |
| Prompt ID | Stable identifier |
| Product mentioned? | Yes/no |
| Recommendation level | 0–3 |
| Position | First, second, third, later |
| Your site cited? | Yes/no |
| Third-party source cited? | URL/domain |
| Sentiment | Positive, neutral, negative |
| Claims accurate? | Yes/no/partly |
| Competitors named | List |
| Notes | Missing feature, wrong price, etc. |

Use this recommendation scale:

- **0:** Not mentioned
- **1:** Mentioned, but not recommended
- **2:** Included in the shortlist
- **3:** Explicitly recommended as a strong or best choice

Then calculate:

### Core metrics

**Mention rate**

```text
Responses mentioning you / total responses
```

**Recommendation rate**

```text
Responses scoring 2 or 3 / total responses
```

**Top-three rate**

```text
Responses placing you in the first three options / total responses
```

**Citation rate**

```text
Web-grounded responses citing your domain / web-grounded responses
```

**Stable recommendation rate**

```text
Prompts where you appeared in at least 2 of 3 runs / total prompts
```

**Recommendation share of voice**

```text
Your recommendation appearances /
recommendation appearances for all tracked vendors
```

Also track an **accuracy rate**. Being recommended with an invented feature, outdated price, or wrong positioning can be worse than not appearing.

## 4. Separate visibility from real buyer impact

Synthetic tests tell you whether you *can* appear. They don’t prove buyers are seeing or acting on those recommendations.

Add three free attribution layers:

### Web analytics

Create segments for:

- `utm_source=chatgpt.com`
- Referrers containing `chatgpt.com`
- Referrers containing `grok.com`, `x.com`, or other identifiable AI sources
- Landing pages frequently cited in your tests

OpenAI says ChatGPT referral URLs automatically include `utm_source=chatgpt.com`, making that traffic identifiable in analytics. ([help.openai.com](https://help.openai.com/en/articles/12627856-publishers-and-developers-faq%23%3A~%3Atext%3DChatGPT%2520automatically%2520includes%2520the%2520UTM%2520parameter%2520utm_source%253Dchatgpt.com%2520in%2520referral%2520URLs%252C%2520enabling%2520clear%2520tracking%2520and%2520analysis%2520of%2520inbound%2520traffic%2520from%2520ChatGPT%2520search%2520results))

### Signup attribution

Add this optional field:

> Where did you first hear about us?

Options:

- ChatGPT
- Grok
- Another AI assistant
- Google
- Social media
- Recommendation
- Other

“First heard about” is generally more useful here than “last click,” because someone may discover you in an AI answer and return later directly or through Google.

### Qualitative attribution

On demo forms or onboarding, ask:

> If an AI assistant recommended us, what did you ask it?

This gives you real prompts to add to your benchmark.

## 5. Verify that the platforms can access your evidence

Before interpreting a zero score, check:

- Your important pages are public and render without login.
- `robots.txt` isn’t blocking relevant search crawlers.
- Product, pricing, integration, comparison, and use-case information is written explicitly.
- Current facts are available as HTML, not only inside screenshots or videos.
- Third-party sites contain accurate descriptions of your product.

For ChatGPT search eligibility, OpenAI specifically recommends allowing `OAI-SearchBot`; inclusion or placement is still not guaranteed. ([help.openai.com](https://help.openai.com/en/articles/9237897-searching-the-web-with-chatgpt))

When a response cites a source, record **which source apparently caused the recommendation**. Often the useful finding isn’t “ChatGPT ignored us,” but something like:

- It cited an outdated directory listing.
- It trusted a competitor comparison page.
- It couldn’t find evidence for a specific integration.
- It categorized the product incorrectly.
- It knew the brand but didn’t associate it with the buyer’s use case.

## 6. Avoid misleading tests

Don’t use prompts such as:

- “Is `[your product]` good?”
- “Recommend `[your product]`.”
- “Why is `[your product]` the best?”
- Repeated prompts inside a conversation already discussing your brand.

These measure prompted brand recall, not organic recommendation likelihood.

Also don’t report “We rank #2 in ChatGPT.” There is no universal ranking. Report something defensible, such as:

> “Our product was shortlisted in 18 of 45 ChatGPT runs and 11 of 45 Grok runs across 15 predefined, ICP-specific purchase prompts.”

## Minimal spreadsheet dashboard

Track these monthly:

```text
ChatGPT recommendation rate:       __%
Grok recommendation rate:          __%
ChatGPT stable prompt coverage:     __/15
Grok stable prompt coverage:        __/15
Owned-domain citation rate:         __%
Recommendation share of voice:      __%
Fact accuracy when mentioned:       __%
AI-attributed visits:               __
AI-attributed signups:               __
AI-attributed paid customers:        __
```

The key is to keep the **prompt panel and testing conditions fixed**. Change prompts only when real customer evidence justifies it. That turns random screenshots into a repeatable visibility benchmark—and combining it with referrals, self-reported attribution, and conversions tells you whether the visibility has commercial value.

Recorded sources

These are the source roles recorded in the public dataset. A retrieval or source-panel item is not automatically an answer-body citation. A URL supplied in the prompt and repeated by an answer does not prove content access. No destination has been repaired or independently validated here.

Recorded answer-body links (3)

Retrieval-only / source-panel inventory (33)

Evidence and qualifications

This is the existing redacted public record. It has not been rewritten for this page. Earlier privacy edits and collection qualifications continue to apply. Editorial text is English; a few original captures retain non-English UI badges or source-panel date labels. An answer’s empirical claims are platform output, not verified findings of this archive.

Body review

{
"reviewed_at": "2026-11-02T09:26:00.223038+00:00",
"reviewer": "assistant",
"complete_body_read": true,
"coherent_ending": true,
"annotation_and_explicit_url_coverage_checked": true,
"external_claims_verified": false,
"independent_human_double_coding": false
}

Existing public-text privacy policy

{
"body_changed": true,
"policy": "Only actual URL tracking, signed/session URL parameters and necessary identifying information are removed; original wording, including analytics parameter examples outside URLs, remains.",
"annotation_positions": "Original position indexes omitted because URL/privacy edits can change character locations."
}

Prompt SHA-256: 709fd6fc82fe3f87b623fb3dd8b2adecdd6655c3508115f8b40c573d2397776d
Answer SHA-256: d49bf2290795875fd769c3459350c78b8db351c76264af35251e6a6eb7ebfe32