Google ranking, brand recognition, and commercial inclusion are three different observations.
Strong Google rankings show that a page performs for particular search queries. A branded AI answer shows that the system can identify your company when your name is supplied. Neither observation establishes that your SaaS enters the considered set when a buyer asks an unbranded question such as “Which tools should I consider for this use case?” The commercial gap therefore has to be measured at the buyer-question level: exact prompt, repeated answers, recommended vendors, explicit preference language, sources, run conditions, and competitor differences. Start there before assuming the cause is technical SEO, backlinks, schema, category positioning, or content. The evidence should determine which branch deserves investigation and whether the gap repeats strongly enough to justify a fix.
Do not collapse three different measurement surfaces into one visibility claim.
Google ranking
A URL performs for a traditional search query. Preserve the query, market, device, page, position range, and observation date before comparing it with another surface.
Branded AI visibility
An answer can identify your company when the prompt supplies its name. That tests recognition and description under those conditions, not whether you enter an unprompted buyer shortlist.
Unbranded commercial recommendation
The answer includes your company when a buyer asks which products fit a category, use case, constraint, or purchasing decision without naming you first.
Why does ChatGPT know our company by name but ignore it when buyers ask for the best tools in our category?
Brand recognition is a different test from commercial inclusion. The useful question for growth leadership is: “Are we entering the considered set when our name is not supplied?” Measure that question directly. If the brand appears but another vendor receives the recommendation, use the focused guide on visibility without selection.
Start with the observed difference, then inspect the next evidence layer.
Branded prompts work, but unbranded category prompts fail.
Evidence to inspect
Inspect whether the answer associates the company with the requested category, buyer, use case, and commercial constraints. Treat those as comparison fields, not disclosed ranking factors.
The same competitors repeatedly enter the shortlist.
Evidence to inspect
Record which vendors are selected for each high-value buyer question and any explicit first-choice language. Compare recurring winners rather than a generic competitor list.
Competitor answers show stronger supporting source evidence.
Evidence to inspect
Compare owned and third-party URLs, the claims they appear to support, and source recency. A displayed citation is evidence to investigate, not proof that it caused the recommendation.
The site ranks for informational queries but commercial inclusion stays weak.
Evidence to inspect
Compare the ranking query and page intent with the unbranded buying question. Check whether current use-case, comparison, fit, pricing, and product evidence actually answers the decision being measured.
Recommendation results change materially between runs.
Evidence to inspect
Repeat under recorded provider, model, prompt, language, location, and time conditions before assigning a durable diagnosis or launching a broad content program.
Preserve enough evidence to compare search strength with unbranded commercial inclusion.
| Evidence | What to retain |
|---|---|
| Unbranded buyer question | Preserve the exact category, use-case, comparison, pricing, or fit wording. |
| Branded counterpart | Where useful, retain the corresponding question that supplies the company or product name. |
| Repeated observations | Keep every controlled run needed to distinguish a recurring gap from one answer. |
| Recommended vendors | Record the complete observed recommendation set and explicit first or preferred choice when present. |
| Answer wording | Retain the relevant response text rather than reducing it to a visibility score. |
| Sources and citations | Save displayed source URLs, the claims they accompany, and when the answer provides no source. |
| Run conditions | Record provider, model, prompt, language, location where relevant, session conditions, and timestamp. |
| Competitor differences | Compare which vendors, claims, and sources recur across the same buyer-question family. |
Treat the gap as a measured commercial problem, not a disclosed ranking formula.
- One missing answer is not total invisibility. It describes one recorded question and run.
- Branded success is not category strength. Supplying the name changes the test.
- A Google ranking is not an AI shortlist guarantee. Measure each surface directly.
- A citation is not a causal explanation. Preserve it as evidence to inspect.
- One model or run is not the whole market. Record scope and repeat before acting.
- A schema or content change is not a promised outcome. Fix supported gaps and retest.
Give the team one evidence-led operating order.
- 01
Freeze the high-value unbranded questions.
Choose the buyer decisions worth winning before measuring visibility.
- 02
Measure recommendation inclusion.
Record whether the brand is absent, included, or explicitly preferred.
- 03
Identify repeated competitor winners.
Find the vendors that recur under the same commercial conditions.
- 04
Inspect answer and source evidence.
Separate observable claims and citations from hypotheses about why they appeared.
- 05
Compare branded and unbranded behavior.
Use the same category and buyer context wherever the comparison remains meaningful.
- 06
Prioritize the highest-confidence commercial gap.
Select a gap that matters to revenue and is supported strongly enough to assign.
- 07
Fix one bounded issue.
Improve the current owned, positioning, product, technical, or third-party evidence you can control.
- 08
Retest.
Repeat the same question family under recorded conditions and preserve the new evidence.
Keep approved commercial questions, retained answers, competitors, sources, and confidence attached.
RecoProof measures approved buyer questions under recorded conditions, retains recommendation and citation evidence, identifies observed competitor displacement, and uses repetition, confidence, and human review before prioritizing a finding. The methodology explains the limits; the fictional sample report shows the evidence-to-action structure.
If an answer includes the brand but describes the product incorrectly, switch to the separate diagnosis for inaccurate or outdated SaaS information.
Current OpenAI documentation used for limited product-behavior claims.
Last checked August 27, 2026. OpenAI documents that ChatGPT Search can return links and citations and that public sites can be eligible for discovery. These sources do not publish a B2B SaaS recommendation formula or establish that one source caused a vendor choice.