Why Your AI Product is Invisible to ChatGPT (and How to Fix It)

You ask ChatGPT for the best product in your category. Three competitors appear and yours does not. Your site is public, indexed, and earning Google traffic, so the result feels irrational.

The useful question is not “How do I optimize everything for AI?” It is “At which step did my product drop out?” Discovery, retrieval, citation, recommendation, and referral are separate events. The fix depends on the first event that failed.

The four outcomes most dashboards blur together

A page request is useful diagnostic evidence. It is not proof that the page shaped an answer. Keep the following outcomes separate:

OpenAI documents three distinct web agents. OAI-SearchBot supports discovery for ChatGPT search. GPTBot controls potential training use. ChatGPT-User is used for some user-triggered requests and is not the automatic search crawler. Allowing OAI-SearchBot makes a page eligible for search discovery, but eligibility does not guarantee a citation or recommendation.

Step 1: Measure the buyer question, not your brand name

Branded prompts answer whether the engine knows you. Buyer prompts answer whether the engine would put you on a prospect's shortlist. Track questions such as:

• “What is the best [category] tool for a small SaaS team?”
• “Which [category] platform has an API?”
• “What are the best alternatives to [competitor]?”
• “Compare [competitor A] and [competitor B] for [use case].”

Use the same prompt set and preserve the engine, date, full answer, recommended brands, answer position, and cited URLs. A single response can change on the next run. A repeated fixed panel shows whether the outcome moved beyond normal answer variance.

Prompt: best [category] tool for a small SaaS team
Engine: ChatGPT
Brand mentioned: yes / no
Recommendation position: 1, 2, 3, or absent
Owned URL cited: exact URL or none
Competitors named: saved as returned
Sources: every cited URL
Run date: timestamp

Step 2: Benchmark the pages that won the answer

Read the cited URLs before editing your own page. Put each source in one of two buckets because the responses require different work.

Bucket A: independent authority won

Review sites, independent comparisons, community threads, and journalist pages can supply corroboration that your own site cannot manufacture. If every selected source is independent, another rewrite of your product page is unlikely to close the whole gap. Give qualified publishers verifiable product facts, offer an unpaid evaluation workspace, and decline paid or guaranteed favorable placement.

Bucket B: a better manufacturer page won

A competing vendor page is actionable. Compare its first screen with yours. Does it answer the exact buyer question? Does it publish current prices, plan boundaries, engine coverage, output fields, limitations, and a visible method? Those checkable facts are easier to use than slogans. Beat the page on clarity and evidence without copying its claims.

This source split matters because the same page will not win every engine. Foglift's quarterly citation benchmark found substantial source divergence across ChatGPT, Claude, Perplexity, Gemini, and Google AI Overview. Track each engine separately instead of reporting one blended citation count.

Step 3: Check discovery and page clarity

If the correct page never appears in the source set, verify the basics: a successful response, an indexable canonical URL, snippet eligibility, relevant internal links, accurate headings, and access for the engine's search crawler. For ChatGPT search, inspect OAI-SearchBot rather than treating GPTBot as the search control.

Next, make the product facts explicit. A buyer and an answer engine should be able to find the following without reconstructing your positioning from several pages:

• what the product is and who it serves;
• which job it completes;
• what evidence or output it returns;
• which plans include the capability;
• current pricing and operating limits;
• important exclusions or uncertainty.

Use valid structured data when it accurately mirrors visible facts. Do not treat it as a recommendation switch. Google's current guidance says there is no special generative-search markup and warns against overfocusing on structured data. Its stated priority is unique, useful, non-commodity content supported by ordinary search fundamentals.

Step 4: Ship one causal change and measure the result

Turn the benchmark into the smallest defensible intervention. Examples include correcting crawler access, replacing a vague opening with a specific product answer, publishing a missing plan boundary, adding an evidence-backed comparison, or pursuing one qualified independent evaluation.

Record the pre-change baseline by engine and landing page. After release, rerun the same prompt set and compare the same outcomes. A crawler fetch is diagnostic. A citation is source selection. A recommendation is shortlist inclusion. A referral is a real person reaching the site. Report all four honestly.

The KDD paper that introduced Generative Engine Optimization tested content interventions on a benchmark of generative-engine queries and reported that effectiveness varied by domain. It did not establish a universal ChatGPT ranking formula, and it did not prove that FAQ schema forces a recommendation. Use research to form a testable change, then rely on your own repeated observations for the verdict.

How Foglift closes the diagnostic loop

Foglift connects four evidence layers. The Technical Audit checks one page across SEO, AI Readiness, performance, security, and accessibility. AI Visibility preserves answer-level mentions, positions, citations, competitors, and sentiment. Deterministic recommendations turn those findings into a ranked fix queue. Sensor analytics keep crawler fetches and human referrals separate.

Every plan includes unlimited single-page Technical Audits. Active Free workspaces receive scheduled weekly Perplexity monitoring. Launch starts at $49 per month and adds all five monitored engines, daily capability with a weekly onboarding default, and API, CLI, and MCP access. The product does not promise that a technical score or crawler visit will become a recommendation.

$ foglift scan https://example.com/pricing --json

# Then inspect the tracked buyer prompt in Foglift:
# mention, recommendation position, cited URLs, competitors, and run date

Frequently asked questions

Start with the first broken link

Run a free Technical Audit on the page that should answer your highest-intent buyer question. Then use the repeatable ChatGPT recommendation workflow to capture the answer and its sources. If you need ongoing evidence, AI Visibility monitoring keeps the five engines, competitors, citations, and outcomes separate.

Sources

• OpenAI crawler documentation: independent roles for OAI-SearchBot, GPTBot, and ChatGPT-User.
• OpenAI crawler documentation: search discovery, user-requested fetches, and potential-training crawler roles.
• Google Search guidance for generative AI features: ordinary search foundations, non-commodity content, and structured-data limits.
• Aggarwal, Murahari, Rajpurohit, Kalyan, Narasimhan, and Deshpande, GEO: Generative Engine Optimization: benchmarked content interventions and domain-dependent effects.
• Foglift Q3 AI Search Citation Benchmark: a reproducible five-engine source-divergence study.

Fundamentals: Learn about GEO (Generative Engine Optimization) and AEO (Answer Engine Optimization) (the two frameworks for optimizing your content for AI search engines).