GEO Lab #03 · S3-CHATGPT-M2-R2 · 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-M2-R2
Platform
OpenAI API + web search
Question / condition
M2
Repeat
2
Model label
gpt-5.6-sol
Visible mode
Not independently recorded
Reasoning
medium
Submitted / first recorded time
2026-11-02T09:28:30.430693+00:00
Completeness
Recorded body complete
Search
Actual search activity observed; exact controls and evidence remain in the raw record.

Original prompt

As a bootstrapped SaaS builder, what's a free, repeatable workflow for checking whether my product appears in ChatGPT and Grok recommendations?

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

A good **$0 workflow** is a small monthly “AI recommendation audit” using ChatGPT, Grok, and a spreadsheet.

## 1. Build a fixed prompt bank

Start with **12–20 prompts** based on how buyers actually shop. Never mention your product’s name in these discovery prompts.

| Prompt type | Template |
|---|---|
| Category | “What are the best `[category]` tools for `[customer type]`?” |
| Use case | “I need to `[job]`. Which tools should I consider?” |
| Constraint | “Recommend affordable `[category]` software for a team of `[size]`.” |
| Alternative | “What are good alternatives to `[major competitor]`?” |
| Comparison | “Compare the leading tools for `[use case]`.” |
| Persona | “What would you recommend to a `[role]` who needs `[outcome]`?” |

Include qualifiers that genuinely differentiate you: bootstrapped, privacy-friendly, self-hosted, no-code, developer-focused, Shopify-specific, under $50/month, and so forth.

## 2. Test two lanes

Run each prompt two ways:

### Lane A: Natural recommendation

Use the prompt exactly as a buyer might:

> What are the best customer-feedback tools for an early-stage B2B SaaS?

This tests whether the model already associates your product with the category.

### Lane B: Web-researched recommendation

Append:

> Search the web, recommend five options, explain why each fits, and cite your sources.

This tests whether your current web presence is discoverable. ChatGPT Search can be explicitly selected and is available on the free tier; Grok’s free plan includes limited real-time web and X search. ([help.openai.com](https://help.openai.com/en/articles/9237897-chatgpt-search))

Don’t ask, “Should you recommend **MyProduct**?” That measures the model’s ability to agree with you, not organic visibility.

## 3. Minimize personalization

For ChatGPT, use an **Unpersonalized Temporary Chat** for every test. This prevents memories and custom instructions from influencing the result. ([help.openai.com](https://help.openai.com/en/articles/8914046-temporary-chat-in-chatgpt))

For Grok:

- Start a fresh chat for each prompt.
- Remove any custom instructions or personalization available in your settings.
- Don’t discuss your brand in that account immediately before testing.
- Keep the account, model, location, and test conditions consistent between audits.

Also record the date and model shown in the interface. Results can change when models or search systems change.

## 4. Run each prompt twice

LLM answers are variable. For a manageable free audit:

- 12 prompts
- 2 platforms
- 2 lanes
- 2 repetitions

That’s **96 answers per audit**. If free limits make that impractical, use one repetition monthly and run the second repetition only for your six highest-value prompts.

Run everything within a short window—ideally one or two days—so the results are reasonably comparable.

## 5. Track results in a spreadsheet

Use these columns:

```text
Date
Platform
Model/mode
Prompt ID
Prompt
Natural or web-researched
Run number
Brand mentioned? Y/N
Position
Top 3? Y/N
Recommended positively? Y/N
Own website cited? Y/N
Third-party source cited
Description accurate? Y/N
Competitors mentioned
Notes
Screenshot/chat URL
```

For `Position`, count only the actual recommendations:

- `1–5`: listed position
- `6`: mentioned outside the main list
- blank: absent

Save the answer or a screenshot. URLs alone can become less useful if interfaces or responses change.

## 6. Calculate four simple metrics

### Recommendation visibility

```text
Answers mentioning you ÷ total tested answers
```

### Top-three rate

```text
Answers placing you in the top 3 ÷ total answers
```

### Researched visibility

Calculate visibility separately for web-researched prompts. This tells you whether the problem is current web discoverability rather than general model awareness.

### Accuracy rate

```text
Accurate descriptions ÷ answers mentioning you
```

A mention with the wrong pricing, positioning, or features is not necessarily a win.

An example dashboard:

| Metric | ChatGPT | Grok |
|---|---:|---:|
| Natural visibility | 17% | 25% |
| Researched visibility | 42% | 33% |
| Top-three rate | 13% | 8% |
| Accuracy | 80% | 50% |
| Own-site citation rate | 20% | 10% |

## 7. Diagnose *why* you are missing

After the blind test, use a separate diagnostic prompt:

> You did not include `[product]`. Search the web and evaluate whether it qualifies for this recommendation. If not, explain the specific evidence or positioning that is missing.

Then ask:

> Which sources most influenced your recommendations?

Treat the explanation as a hypothesis, not an authoritative account of the model’s internal ranking process. Validate it against the actual citations.

Look for recurring gaps:

- Unclear category positioning
- No dedicated use-case pages
- Weak comparison or alternatives pages
- Pricing unavailable publicly
- Few independent reviews
- Outdated product descriptions
- Missing evidence for important claims
- Site blocked from crawling
- Competitors described more concretely

## 8. Check the technical basics

For ChatGPT web visibility, confirm that:

- `robots.txt` does not block `OAI-SearchBot`
- Your CDN or firewall does not block its published traffic
- Important product, pricing, comparison, and documentation pages are publicly accessible
- Pages do not require JavaScript interactions or login to reveal essential facts

Allowing access makes pages eligible; it does **not** guarantee placement. ([help.openai.com](https://help.openai.com/en/articles/9237897-chatgpt-search))

Also monitor analytics for ChatGPT referrals. OpenAI states that referral URLs include `utm_source=chatgpt.com`, making them identifiable in tools such as Google 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))

For Grok, include a few X-centric prompts because its search capabilities can use both real-time web and X information. ([docs.x.ai](https://docs.x.ai/developers/tools/web-search))

## 9. Repeat monthly—not daily

Monthly is usually enough for a bootstrapped product:

1. Run the unchanged core prompt bank.
2. Add at most one or two new prompts.
3. Compare visibility and accuracy with the prior month.
4. Note major website, review, launch, or positioning changes.
5. Investigate only meaningful trends across multiple runs.

Keep **80% of prompts unchanged**. Otherwise, you won’t know whether visibility improved or you merely changed the test.

## Recommended starter version

If you want the leanest possible version:

- Six buyer-intent prompts
- ChatGPT and Grok
- Natural and web-researched lanes
- One run each
- Monthly
- Track mention, position, citation, and accuracy

That’s only **24 answers per month**, takes roughly 30–45 minutes, and gives you a useful directional benchmark. It is not an official “AI ranking”—personalization, location, model updates, search results, and response randomness all matter—but it will reveal whether your product is becoming more or less visible over time.

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 (5)

Retrieval-only / source-panel inventory (29)

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": "2026-11-02T09:29:27.035936+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: 9d96cab49de6d93cb0c1eef3c1da1e5ab5cfdf02f4a703ce57ecf5911dc08219
Answer SHA-256: 51fef7fc3faba4e5fb0219dce5c6e0322f923c6bb7ced1b7cb3e23bf376b5fbb