Aumata AI Visibility Audit: What We Could Verify
Our Aumata AI visibility audit checks citation data, website markup, comparisons, and social discovery. Read the dated evidence and its limits.
An AI visibility audit checks whether a business can be found, cited, and evaluated in AI answers. Our Aumata AI visibility audit found a small amount of recorded citation activity and several gaps in how we make evidence discoverable. It also found that an automated report had missed pages and structured data already on our website.
This is a dated check of our own public website and a third-party citation index, completed on September 7, 2026. It is not a claim that Aumata wins a particular buyer prompt. We have published the source snapshot so you can see exactly what supports the numbers below.
Contents
- Our AI visibility audit results
- What the website check confirmed
- What a citation does and does not prove
- How to make an AI visibility audit repeatable
- The fixes that follow from the evidence
- Frequently asked questions
Our AI visibility audit results
Ahrefs reported four citation links to Aumata across its tracked AI platforms on September 7, 2026. That is a third-party index observation. It is not a count of all answers produced by those platforms, and it is not a count of qualified leads.
We requested the site-explorer/ai-responses-count endpoint for aumata.ai, including subdomains, with no country filter. We requested separate results for ChatGPT, Perplexity, Gemini, Google AI Overviews, Google AI Mode, Copilot, and Grok. The table preserves the citation counts returned by that request.
| Platform | Citation links reported by Ahrefs |
|---|---|
| ChatGPT | 1 |
| Perplexity | 2 |
| Google AI Overviews | 1 |
| Gemini | 0 |
| Google AI Mode | 0 |
| Copilot | 0 |
| Grok | 0 |
| All platforms, as returned by Ahrefs | 4 |
The zeroes mean that this response reported no citation links for those platforms. They do not prove that nobody has ever received an answer mentioning Aumata. A mention without a link, a different prompt, an untracked answer, or a later answer can fall outside this result.
There is also an inconsistency worth showing. The ChatGPT record returns one citation and three pages. We have preserved both fields in the source file. We are not using that page count as a performance claim, and we are not silently changing the response to make the fields agree. It needs clarification from the data provider.
We tried to retrieve the corresponding cited-page records. That request required Brand Radar access that was unavailable on the account used for this check. Consequently, this article does not present prompt text, answer excerpts, or named winning pages. The summary counts are available; the answer-level evidence is not.
That boundary matters. A table with a source and an explicit limitation is useful evidence. A confident claim about the exact buyer questions behind a summary count would be speculation. If your vendor cannot retrieve the underlying answers, its report should make the same distinction.
What the website check confirmed
The website check found working Organization markup, FAQ markup on pricing, and existing comparison pages. An outside report had described these as missing. We checked the returned page HTML and the live URLs instead of treating the report’s labels as verified findings.
The homepage returned Organization, WebSite, Service, and SoftwareApplication JSON-LD. The pricing page returned those types plus FAQPage markup. Both were present in the HTML response, which makes them available without waiting for a visitor to interact with the page.
The comparison pages were also reachable. You can read our marketing agency comparison, in-house marketing comparison, and fractional CMO comparison. Having those pages does not establish that they rank for the queries we care about. It establishes that the problem was not simply a missing page.
| Reported concern | What we could verify | What that leaves to improve |
|---|---|---|
| Missing Organization schema | Present in homepage HTML | Keep company facts accurate |
| Missing FAQ schema | Present on pricing | Keep answers aligned with visible copy |
| Missing comparisons | Existing pages returned successfully | Improve copy and discovery links |
| No owned X account | Connected brand account is @aumatahq | Make the correct profile easy to find |
| No public AI visibility evidence | A dated citation-index response was available | Publish the evidence and its limits |
The social-account check illustrates the same problem. Searching for an assumed handle is different from identifying the account a company actually uses. Aumata’s connected X profile is @aumatahq. The company page on LinkedIn also exists. An account existing does not establish a healthy posting rhythm or useful engagement.
An automated report can still point to a real weakness while getting its diagnosis wrong. If a crawler misses your comparison pages, visitors may also struggle to find them. The sensible response is to inspect internal links and the quality of the pages. Creating duplicate pages because the first set was overlooked would spread the same information across more URLs without solving the discovery problem.
There is another correction to make. Google’s documentation records the retirement of FAQ rich results starting May 7, 2026. FAQPage markup should not be sold as a route to that retired search appearance. Normal questions and answers can still help a buyer understand your service. Read Google’s documentation updates for the current feature status.
What a citation does and does not prove
A citation shows that a source was referenced in an answer captured by the reporting system. A recommendation shows that a business was suggested as an option. Those are different events, and neither one establishes a sale.
Aumata could be cited for an educational article while another firm is recommended as a provider. An answer could also mention our name while citing a third-party website. A domain citation report alone cannot distinguish every one of those cases. You need the actual answer and its context to make that judgment.
This is why we do not turn four recorded citation links into an AI market-share claim. We do not have a defined set of buyer prompts, a denominator of valid answers, or answer-level evidence in this snapshot. We also have no evidence here connecting those citations to bookings or revenue.
Our published customer results answer a different question. They describe work for other businesses, using the sources and periods stated in those case studies. They are not evidence of Aumata’s own citation performance. A provider’s client outcomes and the provider’s ability to market itself should be evaluated separately.
The same care applies to traffic. Third-party search tools estimate traffic from their own coverage and models. Website analytics records visits under its measurement rules. Search Console reports search activity under its own definitions. Those systems can disagree without one number being a direct replacement for another. A report should name the metric before comparing it.
For a founder evaluating marketing, the practical question is what each number lets you decide. A citation record may tell you that a page deserves closer inspection. A qualified booking may tell you that a channel is contributing to pipeline. A rise in impressions may tell you that distribution improved. Treating all three as revenue evidence hides the work still needed.
Our guide to AI citation tracking approaches goes further into choosing a measurement method. The principle behind this audit is narrower: publish what you observed, identify what you could not inspect, and resist filling the gap with a success story.
How to make an AI visibility audit repeatable
A repeatable AI visibility audit keeps the questions, settings, dates, and scoring rules visible. The index snapshot in this article is a starting observation. A proper buyer-prompt benchmark would add a defined sample and saved answers that another person can examine.
Start with questions your buyers ask while deciding what to do. Separate educational questions from provider-selection questions. Someone asking what answer-engine optimization means is at a different stage from someone choosing a marketing team for an established MSP. Combining those questions into one score can hide whether your visibility reaches buyers who are ready to act.
Before running the check, write down the intended market and the question wording. Record the platform, the date, whether web search is enabled, and any relevant location setting. Keep that setup alongside the answer. If you change the question next month, label it as a changed test instead of presenting it as an uninterrupted trend.
For each completed answer, record whether the brand was mentioned, whether the domain was linked, and whether the business was recommended. Save the source URLs and an answer reference or export where available. Review ambiguous cases manually. A passing reference in a warning or an unrelated business with the same name should not become a positive recommendation.
Keep failed checks separate from negative results. A timeout means the check did not finish. A permissions error means the source could not be read. A completed answer that does not name your business is a different observation. Counting all three as absence makes weak infrastructure look like weak marketing.
If you publish a percentage, show the denominator. State how many valid answers were inspected and how many checks failed. Avoid presenting a small sample as an estimate of everything buyers see. Repeating a bounded sample can be useful for your own decisions even when it cannot support a broad market claim.
Then connect the visibility work to business measurement. Tag links in your own distribution, inspect referral data where available, and ask prospects how they found you. Do not assume an AI citation produced a booking because both occurred in the same month. The attribution question needs evidence of its own.
For this article, we have not completed that prompt benchmark. We have completed a website check and retrieved an index snapshot. That is the baseline readers should use when judging the claims here. A later update should add a new dated artifact, preserve this one, and explain any changes in the measurement method.
The fixes that follow from the evidence
The immediate fixes are about making useful information easier to find and easier to evaluate. They do not depend on promising a particular ranking or an AI recommendation. They improve what a buyer can inspect before speaking with us.
First, make the existing comparisons visible from the homepage and pricing page. A buyer deciding between an agency, an employee, and a fractional CMO needs a clear route to those tradeoffs. The comparison hub brings those options together, including situations where another approach fits better.
Second, remove unsupported generalizations from comparison copy. Agencies do not all have the same contract. In-house marketers do not all have the same salary or seniority. Fractional CMOs do not all exclude delivery. A useful comparison asks who owns the work, what the scope includes, what costs sit outside it, and how the engagement ends.
Third, distinguish the base engagement from added channels. Our pricing page defines the current starting scope. Ads, LinkedIn, and outbound are scoped separately when the plan calls for them. A headline price should not imply that every channel is included at that floor.
Fourth, link the official social profiles where visitors can find them. We should not expect a buyer to guess the handle or trust an unverified account with a similar name. Visible profile links and accurate company markup should describe the same organization.
Fifth, give proof a stable URL. This article and its source file make our own citation observation inspectable. Named customer stories require the customer’s permission and a supported account of the work. Anonymous case studies can still have stable links, useful methodology, and honest limitations while that permission is being arranged.
Finally, keep distribution separate from evidence creation. A review must come from a real customer describing their experience. A community contribution should answer the discussion and disclose the writer’s affiliation. A product listing should describe a tool someone can use. Publishing these assets can create opportunities to be found, but their existence is not proof of adoption or results.
Frequently asked questions
Does Aumata appear in AI answers?
Ahrefs reported four citation links across its tracked platforms in the September 7, 2026 snapshot. We have not retrieved the underlying answers for this article, so we cannot claim which buyer prompts recommended Aumata. The raw response and its limitations are linked above.
Does a zero in the table mean a platform never mentions Aumata?
No. It means the requested snapshot reported zero citation links for that platform. The reporting system does not represent every answer, every prompt, or every mention without a link.
Was the outside audit wrong about everything?
No. It identified useful concerns about distribution and public proof. But its claims about missing schema, missing comparisons, and the lack of an owned X account did not match the checks described here. Findings need individual verification.
Can schema guarantee AI citations or rich results?
No. Structured data describes your content; it does not prove that an answer engine selected it. Google also retired FAQ rich results in May 2026. Keep useful answers and accurate markup, and measure visibility separately.
If you want to see what a buyer can currently find about your business, start with the same evidence-first approach. Run the free audit on your site, talk to a strategist, or see pricing.