An AI answer can cite your website as a source while recommending other businesses. The citation shows that your content was referenced in that answer. It does not automatically mean your company was presented as a suitable provider. A Dubai business measuring GEO should record both observations separately.
For example, an answer may use an agency's pricing guide to explain typical cost factors, then name different agencies in its shortlist. This is an illustrative scenario, not a report of a particular client's results. It shows why a citation total alone cannot answer whether buyers are being directed towards your business.
Classify what the answer actually does
| Observation | What to record | What you can conclude |
|---|---|---|
| Your URL is cited | Exact page and the statement it supports | That page was used as a visible source |
| Your brand is mentioned | The wording and context | The business appeared, with the stated description |
| Your business is recommended | The buyer question and reason given | It was presented as an option for that particular need |
| A visitor arrives | Available referral and landing-page evidence | A measurable visit occurred under the tracking limits |
| A suitable enquiry follows | The actual enquiry and source evidence | A commercial interaction occurred, with attribution qualified as needed |
Microsoft's AI Performance documentation describes citation counts and notes that page-level citation activity does not establish ranking or placement. Our recommendation distinction goes a step further by reading the answer itself: did it actually suggest the business for the buyer's need? You can apply this classification in your existing internal measurement system.
Read the answer as a buyer would
Suppose someone asks which Dubai agency can produce presenter-led videos for a service business. The answer may name several providers, explain what to compare and cite a separate guide about production costs. If your guide supplies the cost explanation but your business is absent from the provider options, you have earned source visibility for that answer. You have not yet established supplier visibility for the same question.
Now consider a different answer that names your business but gives an inaccurate service description. A simple brand-mention counter may record a success, while a buyer could receive the wrong expectation. This is why the useful review includes the role of the mention and the accuracy of the explanation. Counting appearances without reading them can hide the very issue the business needs to fix.
Use a short classification guide for the people reviewing answers. A recommendation means the answer presents the business as an option for the stated need. A passing mention, a warning, a comparison reference and a source credit should not automatically receive that label. Preserve the relevant sentence so another reviewer can understand why the classification was chosen.
Define the denominator before reporting a percentage
A percentage needs a clear base. Are you counting answers, unique questions, visible citations or test runs? These are different units. One answer can cite several pages from the same business, and one question can be tested repeatedly. A report that changes the denominator between months can show a movement that comes from the measurement method rather than from a change in visibility.
For a simple answer-level measure, mark whether each eligible answer contains at least one citation to the business’s website. Count the answer once for that measure, even if it contains three such links. Separately record whether the answer recommends the business. Keep the raw links for page-level analysis. This is a proposed reporting convention, not a claim that every platform uses the same definition.
State how failed or empty runs are handled. An unavailable response should not silently become a negative recommendation observation, nor should it disappear without explanation. Keep a count of eligible answers and excluded runs with their reasons. The business can then compare like with like and see whether coverage changed.
A worked example with separate citation and recommendation counts
Imagine a fictional review of 40 eligible answers to a fixed group of Dubai service-provider questions. The business website is cited in 12 answers and the business is recommended in five. Three answers contain both. The figures below are invented to explain the method; they are not LunaSol performance data. They show why adding citation and recommendation counts together would double-count some observations.
| Observation | Answers | Share of the 40 eligible answers |
|---|---|---|
| Website cited at least once | 12 | 30% |
| Business recommended | 5 | 12.5% |
| Both cited and recommended | 3 | 7.5% |
| Cited without a recommendation | 9 | 22.5% |
| Recommended without a website citation | 2 | 5% |
| Either cited or recommended | 14 | 35% |
The nine citation-only answers deserve a content review: what information did the pages supply, and was the question actually asking for suppliers? The two recommendation-only answers deserve a different review: how was the business described, and what visible evidence supported its inclusion? Neither group is automatically a failure. The interpretation depends on the question and the commercial objective.
Do not describe the 35% combined figure as a recommendation rate. It measures a broader condition that includes source use without supplier selection. If the company’s main goal is to appear as a suitable agency for commercial buyer questions, keep the recommendation result visible on its own. A combined metric can be supplementary, but its name should explain exactly what it counts.
Segment the results by the job each question performs
Group questions into practical buyer stages: understanding a problem, comparing approaches, selecting a supplier and checking a named business. A general explanation may naturally make more use of educational sources. A supplier-selection question creates a clearer opportunity to observe recommendations. A branded verification question tests whether the answer accurately understands a business the user already knows.
Keep these groups visible rather than mixing a large set of easy branded questions into a small set of difficult unbranded questions. The overall percentage can otherwise improve while the business remains absent from the commercial searches it cares about. Also preserve language, location context and service category so the team can identify where the actual gap sits.
Improve the route from a useful source to a credible offer
When an article earns citations, inspect whether a human reader can identify who published it and where the relevant service is explained. A cost guide can remain neutral and useful while linking naturally to a service page for buyers who want help. The link should connect the subject of the article to a real offer. Adding unrelated calls to action throughout the text can make the guide harder to use.
Then assess the destination independently. Does it explain the buyer fit, scope, process and relevant work? Does it resolve the question that prompted the visit? A strong article cannot repair a service page that makes vague claims or sends enquiries to the wrong team. Treat the source article and commercial destination as connected pages with different jobs, and review both sides of that connection.
For LunaSol, a guide about preparing presenter-led production should lead to the actual video service, where a buyer can inspect the offer and public examples. The guide should still teach the preparation process fully. This creates a useful customer path without pretending that every source citation is already a sales recommendation.
Review changes without claiming a causal result too early
Keep the question set and classification rules stable across the comparison. Note the release dates of content changes, but also record other relevant changes in the business or test setup. If recommendations increase after a service-page update, that timing is worth investigating. It does not, by itself, prove the update caused the increase. Read the answers and inspect whether their explanations changed in the way the page was intended to support.
Use the result to decide the next action. Repeated citation of a clear explanation may justify strengthening that subject with useful evidence. Repeated incorrect descriptions call for a factual-source investigation. Persistent absence from well-matched supplier questions may call for a closer review of offer clarity and public proof. These are different editorial tasks, so the report should not reduce them to “publish more blogs”.
Before applying the classification across a large archive, have two reviewers independently label a small shared sample. Compare disagreements and refine the definitions. For example, “you could also research Company A” may need a consistent rule about whether it counts as a supplier option. Keep that rule with the report. This makes later comparisons more reliable and prevents a change in reviewer interpretation from masquerading as a change in the business’s visibility.
Questions about citation and recommendation reporting
Is a citation still valuable if the business is not recommended?
It can be. The content contributed to the answer, and a visible source may offer a route to the site. Its value depends on the question, the cited page and what happens next. Record it accurately as source visibility and evaluate commercial outcomes separately rather than dismissing it or overstating it.
Do we need another tracking platform to make this distinction?
No, if the existing system preserves the question, answer and visible sources. Add clear classification fields and review the underlying responses. The important work is consistent interpretation and a stable comparison, not the purchase of another dashboard. A small, carefully reviewed sample can expose problems that a large unexplained total hides.
Inspect the page that earns the citation
Ask what the page helps the answer explain. A factual cost study may be useful evidence for a pricing question. That does not necessarily establish the author's service suitability. Check whether the page identifies the business clearly, links to the relevant service and explains the relationship between the research and the work offered.
Preserve the information that makes the page useful. Do not turn every paragraph into a sales pitch or remove neutral comparisons merely to insert your brand more often. Add a relevant route for a buyer who wants the service, while keeping the evidence understandable and accurately sourced.
Strengthen the commercial page independently
The service page should explain who the work suits, what is delivered, the real business location or coverage and the evidence supporting capability. A buyer should be able to evaluate the offer after following the link. Supporting project pages and accurate public business information can make that evaluation easier.
Measure the intended outcome separately
Track citation and recommendation rates across the same defined question set, languages and engines. Keep branded tests separate from general provider questions. Our website citation guide covers source visibility; a discussion through our GEO page can focus on the gap between being used as a reference and being considered for the service.