ChatGPT can link to your website without recommending your product. A citation tells you that your page appears as a source. A recommendation means the answer presents your product as a suitable choice for the buyer. Those are different outcomes—and fixing the gap starts with identifying which one is missing.
For a SaaS company, the useful question is: “Are we being used to explain the problem, or being considered as a solution?”
This guide gives you a repeatable way to investigate that question, decide which page to improve, and measure what changes.
A citation, a mention, and a recommendation mean different things
Use three separate labels when reviewing an answer:
- Citation: your domain appears in an attached source link. Record the exact URL.
- Brand mention: the answer’s main text explicitly names your company or product. A name that appears only inside a source card does not count under this definition.
- Recommendation: the answer presents your product as a suitable option for the stated need. Record whether the endorsement is conditional and whether any limitations are mentioned.
A neutral list entry may be a mention without a recommendation. A criticism is also a mention. Neither should automatically be reported as a positive result.
OpenAI explains that search responses may include citations and that its Sources panel can contain both cited sources and other relevant links. For a stricter audit, record an attached citation separately from a link that appears only in that panel. OpenAI’s guide to ChatGPT Search.
Illustrative example: imagine a fictional scheduling product called Acme Scheduler. ChatGPT links to its guide about reducing missed appointments but recommends another tool for a multi-location business.
Acme Scheduler has earned a source link. That answer alone does not establish whether its product suits the buyer—or explain why it was not selected.
What research tells us—and what it doesn’t
The gap between being cited and being named is documented. Semrush and Growth Memo studied 3,981 domain appearances across 115 prompts, 14 countries, and four AI search engines. In the results breakdown, 61.7% of all recorded appearances were citations without a brand mention. That denominator matters: this is not a universal rate for every ChatGPT citation. Semrush’s original study.
A separate BuzzStream study analyzed 12,000 responses across four platforms, approximately 200 brands, and three prompt types. It reported that 23.1% of brand mentions also had a citation to the brand’s domain in the same response. This is a different sample and measure. It should not be combined with the Semrush number into a single benchmark. BuzzStream’s methodology and findings.
These studies support measuring the outcomes separately. They do not prove why a particular product was omitted, or that any single edit will earn a recommendation.
Three explanations worth testing
Treat these as working hypotheses. The evidence you collect should determine which one applies.
1. Your page answers an informational question
A guide explaining appointment reminders can be useful even when the reader is not shopping for scheduling software. If the prompt asks how reminders work, a product recommendation may be unnecessary.
Check: classify the original prompt as informational, comparison, or buying intent. A missing recommendation on an informational prompt is not automatically a problem.
2. The product’s fit is hard to verify
Your educational content may be strong while the product page leaves important questions unanswered:
- Does the product support the buyer’s team size?
- Is the required integration available?
- Which plan includes the relevant feature?
- What setup work is required?
- Where are the limitations documented?
Check: use the exact requirements in the prompt to review your public product pages. Mark each requirement as supported, contradicted, or unknown. An unknown is a content gap to investigate, not permission to invent a capability.
3. The sources describe your company inconsistently
Your homepage might position the product for small teams while an older directory profile describes it as an enterprise service. That discrepancy deserves attention even if you cannot prove it caused the answer.
Check: open the sources attached to the response. Identify factual errors, outdated descriptions, and missing evidence relevant to the buyer. Do not assume that every third-party mention would be useful.
How to audit the gap in one working session
This is a suggested starting method, not a statistically representative sample of everyone using ChatGPT.
Step 1: Choose a small, fixed prompt set
Start with ten questions drawn from sales calls, support requests, or customer interviews:
- Three questions about the problem your product solves.
- Four questions asking for suitable tools or vendors.
- Three questions about comparisons or specific requirements.
Keep branded and non-branded prompts in separate groups. Asking about your company by name makes it unsuitable for measuring whether buyers would discover you unaided.
For fictional Acme Scheduler, a non-branded buying prompt could be:
Which appointment scheduling tools support a small business with three locations and SMS reminders?
Do not add the brand name to that prompt to improve the result.
Step 2: Record the test conditions
Write down the date, platform, model if shown, language, market, and whether web search was used. Keep the settings consistent when repeating the audit.
OpenAI documents that ChatGPT can rewrite a question into more targeted search queries and can use location or saved memories to inform search. Those are reasons to record your test context. How ChatGPT searches the web.
Keep API-based monitoring and tests in the ChatGPT app labeled separately. They are different testing environments; do not present one as an exact reproduction of the other.
Step 3: Save the answer and inspect the links
For each answer, record:
- Whether your domain is attached as a citation.
- Whether your brand is named in the main answer.
- Whether your product is recommended for the requested use case.
- Which competitors appear.
- The exact cited page and the claim it supports.
- Any factual error or missing requirement.
Open each relevant citation. A link’s presence does not establish that the page supports the claim beside it.
Avoid treating a follow-up answer to “Why didn’t you recommend us?” as a reliable explanation of the earlier selection process. Use the actual response, linked pages, and verifiable product facts as your evidence.
Step 4: Repeat the sample
As a manageable baseline, run the ten prompts three times in fresh conversations under consistent conditions. That gives you 30 observations.
Keep failed runs in the log. Decide beforehand how you will count completed answers, refusals, and answers without citations. If a run did not use the required search setting, label it instead of quietly combining it with the others.
For rate calculations, report the number of valid observations and exclusions. Repeated prompts are not 30 independent buyers, so the resulting percentages describe your test—not market-wide visibility.
Step 5: Choose one gap to address
Prioritize a page when all three conditions hold:
- The prompt reflects a real buying situation.
- Your product actually meets that buyer’s requirements.
- Public evidence of that fit is missing, unclear, or outdated.
If your product does not fit, improve the targeting or the product. Rewriting a page cannot honestly remove a real limitation.
What to change on a cited page
Begin with the page and use case you identified. The following are editorial improvements to test, not confirmed ChatGPT ranking factors.
Connect the explanation to a real capability
A vague sentence gives a buyer little to verify:
Our platform helps teams work smarter.
A more useful sentence is specific:
Acme Scheduler lets a business manage appointment availability across three locations from one account.
That is a fictional example. Publish a statement like it only if the capability is real, and link to the documentation that explains it.
Show the conditions and limitations
If an integration requires a paid plan, say which plan. If setup takes several steps, describe them. If a workflow is unsupported, disclose that.
A clear product page should help someone decide whether the product fits, including when the answer is no.
Add evidence that someone can inspect
Useful additions include an annotated screenshot of a real workflow, an implementation walkthrough, or a customer example published with permission.
Google’s guidance encourages original information and substantial value beyond rewriting existing sources. A concrete example from your product is a stronger editorial contribution than another generic list of “AI SEO tips.” Google’s helpful-content guidance.
Make the next page easy to find
Link the relevant educational section to the specific feature or use-case page. A reader researching reminders should reach the reminder documentation without searching your navigation.
For Google AI Overviews and AI Mode, Google says ordinary SEO foundations remain relevant, including internal links and accessible text. It does not require special AI markup. That guidance concerns Google’s features; it is not a promise about ChatGPT recommendations. Google’s AI features documentation.
Measure progress without inflating the result
For a single brand, count each outcome at most once per valid response:
- Citation rate: responses with a citation to your domain ÷ valid responses.
- Mention rate: responses naming your brand in the main text ÷ valid responses.
- Recommendation rate: responses recommending your product ÷ valid responses.
Multiply each result by 100 to express it as a percentage.
In a fictional 30-response example, suppose the website is cited in 12 answers, the brand is named in six, and the product is recommended in three.
That is a 40% citation rate, a 20% mention rate, and a 10% recommendation rate. The groups can overlap, so do not add the percentages together.
Record how many of the cited answers also name or recommend the brand. Those intersections tell you whether citations and recommendations occur in the same answers.
For definitions and sampling guidance, see our AI visibility score guide. If competitors repeatedly appear instead, use the competitor mention tracking workflow.
Your next four weeks
Use this as a practical schedule, not a promise of results within 30 days.
Week 1: establish the baseline and choose one buying-intent gap.
Week 2: publish a focused improvement with accurate product evidence. Record what changed and when.
Week 3: repeat the same prompts under the same conditions. Preserve unfavorable answers as carefully as favorable ones.
Week 4: compare citation, mention, and recommendation counts. Review whether any new recommendations are accurate and relevant.
An increase after an edit is an observation, not proof that the edit caused it. If you need stronger evidence, repeat the measurement over more periods and track a comparable set of pages you did not change.
Frequently asked questions
Is a citation without a brand mention worthless?
No. It still provides a link readers may follow. Whether it helps the business depends on the audience, the cited page, and what visitors do next. Measure referral visits and leads separately from answer visibility.
Should every cited article recommend its publisher’s product?
No. An informational answer can be complete without naming a vendor. Focus on buyer questions where your product is a credible fit.
Will adding the brand name more often solve this?
There is no reliable guarantee. Make the company’s role and relevant capabilities clear where they belong. Repetitive brand stuffing does not supply missing evidence.
How long until ChatGPT reflects a page update?
This workflow cannot promise a refresh or recommendation deadline. Date your changes and rerun the same test instead of assuming the next response will incorporate them.
Start with one answer you want to improve
Choose a real buying question, save the answer, and identify whether the missing outcome is a citation, a mention, or a recommendation.
Then improve one specific piece of evidence that a buyer could verify. That gives you a focused experiment and a clearer reason for publishing the next page.
Explore BotSeen’s diagnosis workflow to see how it helps organize AI visibility findings and suggested fixes.