Your SaaS has visibility, but the answer does not select it for the buyer's need.
A mention or citation shows that your brand or content is present in one observed answer. A recommendation is a different commercial signal: the answer presents a vendor as a plausible choice for the stated use case. Before changing pages, verify the gap across the exact buyer question, repeated observations, recommended vendors, answer language, and displayed sources. Then compare the evidence associated with your brand and the selected competitors. The issue may be category fit, commercial proof, current product information, third-party framing, or simply an unstable result. The evidence should decide which branch deserves investigation.
Why ChatGPT recommends your competitor: six reasons to check
No AI company publishes how it chooses vendors, so treat these as hypotheses to test, not proven causes. Each one comes with the check that confirms or rules it out. The figures are from our study of 768 citations in 64 AI answers to software buying questions, collected through the APIs of an OpenAI model with web search and Perplexity Sonar. API answers can differ from what the ChatGPT app shows a signed-in user.
- Reason 01
The pages it reads list your competitor, not you.
What we saw
85% of the 768 cited URLs in our study were third-party pages such as comparison articles, "best X" lists and review blogs. In the email marketing questions the most-cited domains were sequenzy.com, ventureharbour.com and encharge.io, not any of the five vendors we tracked.
How to check
Open every source cited for the question you lose. Count how many name your competitor and how many name you. If you are missing from the lists, that is the gap to work on first.
- Reason 02
Your own website is a small part of the evidence.
What we saw
The tracked vendors' own sites supplied 11% of the cited URLs (81 of 768). A better homepage helps, but it is rarely the only page the answer is built from.
How to check
Note which of the cited pages you control. If none are yours, improving your site alone is unlikely to change the answer.
- Reason 03
The competitor says who it is for; you list features.
What we saw
Buyer questions carry a constraint: team size, industry, budget, an integration. Answers pick vendors that match it. A competitor whose pages say "for agencies" or "for 10-person teams" is easier to match than one that lists capabilities.
How to check
Write down the constraint in the buyer question. Then look for that exact fit on your homepage, pricing page and the third-party lists from the first check.
- Reason 04
Community threads favour them.
What we saw
Perplexity Sonar cited Reddit in 28 of its 32 answers; the OpenAI model with web search cited it in none of 32. Where an engine reads forums, what users recommend there travels into the answer.
How to check
Search Reddit and community forums for the same buyer question. Note which vendors people recommend and why.
- Reason 05
ChatGPT files you under a different category.
What we saw
If the model thinks you are a CRM when you sell sales engagement, you will not be recommended for sales engagement questions however good your pages are.
How to check
Ask the branded question: "What is [your brand] and who is it for?" If the description is wrong or out of date, fix that before anything else.
- Reason 06
The result is not stable yet.
What we saw
On a rerun of the same question, the OpenAI model with web search cited the same set of domains in 2 of 16 pairs; Perplexity Sonar in 14 of 16. One losing answer can be a different draw, not a pattern.
How to check
Ask the same question at least three times, on different days. If the competitor wins only some of the runs, measure before you change content.
Mention, citation, and recommendation answer different questions.
Mention
The answer names your company or product. It proves presence in that observed answer, not buyer-fit endorsement.
Citation
The answer links to or identifies a source. The source may support part of the response, but its appearance does not prove it caused the vendor choice.
Recommendation
The answer presents a vendor as a plausible choice for the buyer's stated need. Explicit first-choice or preferred-fit language is stronger than inclusion in a long list.
OpenAI documents that ChatGPT Search can search the web and present links to relevant sources. That supports inspecting citations as answer evidence. It does not provide a public causal explanation for why one B2B SaaS vendor was recommended over another.
Start with the observed pattern, then inspect the next evidence layer.
Your brand is mentioned, but it is not in the shortlist.
Evidence to inspect
Check whether the answer associates you with the requested category, use case, company size, constraints, and commercial intent. Then compare the specific fit evidence stated for shortlisted competitors.
Your site is cited, but another vendor is recommended.
Evidence to inspect
Inspect what claim the citation actually supports. Your page may supply a definition or market fact while the vendor comparison relies on different evidence. Do not infer a causal path from citation to selection.
You appear in the shortlist, but a competitor receives preferred-fit language.
Evidence to inspect
Compare the buyer criteria named in the answer—such as product scope, target market, integrations, pricing clarity, or proof—and verify those claims against current owned and third-party sources.
The winner changes across repeated runs.
Evidence to inspect
Treat the result as variable until the prompt, model, time, location or context, answer text, and source behavior are recorded. The next action may be better measurement rather than an immediate content change.
Build a reviewable record before assigning content, SEO, or product marketing work.
| Evidence | What to retain |
|---|---|
| Exact prompt | Preserve the buyer wording, constraints, and whether the query is branded or unbranded. |
| Run conditions | Record provider, model, date and time, location or language where applicable, and relevant session context. |
| Repeat observations | Run enough controlled checks to distinguish a recurring pattern from one answer. |
| Recommendation set | Capture every vendor presented as a plausible choice, plus any explicit first-choice or preferred-fit language. |
| Raw answer | Retain the relevant answer text instead of reducing the result to a score or screenshot. |
| Sources and citations | Save the displayed source URLs and the claims they appear to support; note when the answer shows none. |
| Comparison evidence | Check current owned pages and credible third-party descriptions for the target brand and selected competitors. |
A missing recommendation and a wrong product description need different evidence.
If branded prompts identify your company but unbranded category questions omit it, use the guide to diagnose why you can rank on Google without entering ChatGPT recommendations. If the answer includes your company but gets pricing, capabilities, positioning, or integrations wrong, follow the separate workflow for an inaccurate or outdated AI description.
Visibility without selection is a diagnosis prompt, not proof of a cause.
- One screenshot is not a trend. A single answer cannot establish a stable recommendation gap.
- A citation is not a recommendation. Your page can be used as evidence while another vendor is selected.
- A citation does not prove causality. A displayed source does not reveal the complete mechanism behind vendor selection.
- One omission is not total invisibility. The result applies to the recorded question and conditions.
- Competitor presence is not causal displacement. It shows that another vendor appears in the observed recommendation set.
- A content change is not a guaranteed correction. Retest rather than promising an outcome.
Prioritize the smallest supported gap that could change a real buying decision.
- 01
Verify the commercial recommendation gap.
Confirm that the prompt represents a real buying decision and that the brand is present without being selected.
- 02
Find the affected buyer-question family.
Determine whether the pattern is isolated or appears across related category, comparison, use-case, pricing, or fit questions.
- 03
Compare the vendors actually recommended.
Use the observed recommendation set, not a generic market competitor list.
- 04
Inspect the answer and source differences.
Separate claims made in the answer from owned and third-party evidence that can be checked.
- 05
Choose the highest-confidence controllable gap.
Prioritize current, commercially important evidence that is missing, unclear, inaccurate, technically inaccessible, or weak relative to the buyer criteria in the answer.
- 06
Make one bounded change and retest.
Define an owner and acceptance criteria, publish the change, then repeat the same question family under recorded conditions.
A cited definition does not put the cited vendor on the shortlist.
Imagine a buyer asks for a reporting platform for a 50-person SaaS team. The answer cites Vendor A's glossary for a definition, mentions Vendor A in passing, and recommends Vendors B and C because the answer describes their team fit and reporting workflow. The useful next step is not “get more citations.” It is to verify whether that pattern repeats, inspect the fit claims made for B and C, and check whether accurate, current evidence about Vendor A's team fit is available on owned and credible third-party pages. This is illustrative, not a RecoProof customer result.
Keep the commercial question, raw answer, competitor set, and sources attached.
The paid Recommendation Audit tests approved commercial questions under recorded provider conditions. It retains the answer and citation evidence, separates mentions from recommendations, compares three competitors, applies confidence and human review, and turns supported gaps into a prioritized 90-day action plan. It does not claim that an individual page caused a recommendation or promise that a change will produce future inclusion.
Review the measurement method and the explicitly fictional sample report before buying. If the gap is not yet confirmed, start with the bounded free AI visibility check.
Current OpenAI documentation used for product-behavior claims.
Last checked October 29, 2026. OpenAI product behavior can change; these sources support the limited claims about web search, links, citations, and publisher discovery above—not a theory of B2B vendor ranking factors.