Audit Citation Presence for French Queries

Before this lecture, you should understand from Lecture 1 why a cited answer is a different object from a ranking report. You should also carry forward Lecture 2’s habit of reading source selection through reachability, usefulness, freshness signals and authority signals before recommending changes.

The first useful audit I ever ask a student to build is ugly. Not strategic-looking. Not a dashboard with a dark background and smooth curves. It is usually a sheet with twelve rows, two languages, four columns that keep getting widened, and a few comments that look almost childish: “client absent,” “directory cited,” “wrong category,” “English page used for French question.” In a Dutch agency team, someone will try to clean it up too early. I usually stop them.

A composite scenario will do. A French-market B2B software client asks why Perplexity mentions a competitor for “meilleur logiciel de gestion des demandes fournisseurs,” while its own page ranks reasonably well in ordinary search. The agency runs the same idea three ways: in French, in English, and in a mixed query where the category stays English but the market is French. The answer changes more than expected. In one row, the client is absent. In another, the client appears but is described as a generic purchasing tool. In the mixed query, Perplexity cites an English partner page and skips the French service page. The awkward row is not a failure of the audit. It is the audit beginning to speak.

From single checks to a citation audit

A citation audit is a review of queries, answers and cited sources for client representation, because Perplexity visibility only exists in the relation between the prompt, the answer and the evidence. One answer can mislead you. A small set of related answers starts to show a pattern.

The temptation is to type one dramatic query, see that the client is missing, and walk into the client call with a diagnosis. That is too fast. Perplexity’s answer-and-cite model reacts to the wording of the question, the language used, the implied buyer, and the source material available at that moment. A single check tells you what happened once. A citation audit asks whether the client is consistently absent, inconsistently described, or cited only through sources the agency would rather not rely on.

In Lecture 1, we named citation presence as the appearance of a client, page or source as evidence in a Perplexity answer. Lecture 3 gives that idea a working shape. We stop saying “Perplexity does not see the client” as if the system had one stable opinion. We ask, “For which French-market questions does the client appear, which source supports the appearance, and what wording does the answer attach to the business?”

That wording matters. A French industrial SaaS company may appear in an answer but lose its real category. Perplexity might call it a procurement platform when the page positions it as supplier request workflow software. A client team may hear “we were mentioned” and feel relieved. The agency should be less easily soothed. Citation presence without accurate representation is like being invited to a meeting under the wrong job title.

A good audit therefore records three objects together: the query, the answer wording, and the citations. Keep them joined. If you separate them too early, you will later forget which source supported which claim.

Build the first query set like a buyer, not a keyword collector

The first query set should be small enough to read slowly. For most agency training exercises, I prefer eight to twelve questions. Fewer than that and one odd answer can dominate the interpretation. Much more than that and the student starts sorting instead of reading. We are not making a keyword universe here. We are building a controlled window into Perplexity’s evidence behavior.

Start with the client’s actual commercial situations. For a French-market B2B client, that usually means a few category questions, a few comparison questions, and a few problem-led questions. A category question might ask for software providers in France. A comparison question may ask which tools suit industrial procurement teams. A problem-led question may describe a buyer pain without naming the category. Each type pulls different sources into view.

A query log is a record of prompts, languages, answers, citations and mismatches. I want the word “record” to feel a little heavy. The log is not a pile of screenshots. It is a structured memory of what the system said under specific conditions.

For Dutch agencies working across French and English material, language choice belongs in the log from the first row. Do not only translate the same query mechanically. Translation changes the object. “Supplier request management” and “gestion des demandes fournisseurs” can point to overlapping but not identical commercial meanings. A mixed-language query, such as an English category with “France” or “marché français,” may reveal which evidence Perplexity finds easier to use. That does not yet prove why the system made the choice. It gives you a clue worth keeping.

A teaching example: imagine a client that sells to French manufacturing groups but keeps its strongest product explanation in English documentation. In French, the query returns trade directories and a competitor comparison page. In English, the answer includes the client but cites a partner page. In the mixed query, Perplexity names the client and adds one slightly wrong limitation about company size. The row is messy. Good. Real audits have crumbs on the table.

The first query set should also include one or two queries where you expect the client not to appear. This feels backward, but it protects the audit from vanity. If the company only serves industrial procurement teams, do not test it only with broad “best B2B software France” questions and act wounded when it disappears. A scope-check row helps you separate a genuine missing-citation problem from a query that was never close enough.

Read source types before judging the client page

Once the rows are filled, most people look for a single culprit. The client page is weak. The competitor is stronger. The directory is stealing visibility. Sometimes one of those statements is partly true. Usually the cited set is more layered.

Read the source types first. A trade directory may support the company category. A partner page may support a relationship. A client page may support capability if it states the capability cleanly. A comparison article may support the market frame. In Lecture 2, we looked at source selection. Here, the audit records which source type keeps appearing for which kind of query.

Object A, our composite French B2B SaaS firm, becomes useful again under a different light. Earlier, we used it to see why a thin directory could be selected. In this lecture, the same kind of client is placed into a query log. The agency runs ten French-market questions. The directory appears in four. The client’s own page appears in one. A competitor page appears in three, but only on comparison-style questions. Two answers cite no commercially useful source at all. One answer names the client but attaches an old category from a third-party profile. No single row explains the case. The pattern across rows does.

This is where the audit becomes more useful than a complaint. If directories appear only for category questions, the agency may need to inspect whether the client’s own page states the category clearly enough. If partner pages appear for integration questions, those sources may be doing a narrow job well. If competitor pages appear for problem-led questions, the client may lack usable evidence for that buyer problem. We are still diagnosing; rewriting comes later.

Be careful with emotional language in the notes. “Bad source” is rarely useful. Write what the source does. “Cites directory for category.” “Uses partner page for market confirmation.” “Client named but own site absent.” “French answer leans on English source.” These short notes look plain, but plain notes survive client calls better than clever labels.

A recurrent pattern in French audits is that Perplexity does not ignore the client completely; it uses a weaker public version of the client. That is more troublesome than absence. Absence says the evidence path did not reach the business. A distorted appearance says the system found something, but the available source shape nudged the answer sideways.

Catch language shifts without over-explaining them

French-market audits need language discipline. The agency may be Dutch, the client may sell in France, the product team may write in English, and the web may contain half-maintained profiles in both languages. Perplexity can move through that material in ways that look sensible in the answer and odd in the citations.

Do not invent certainty where the evidence is thin. If a French query cites English material, you can record that fact. You can compare the English source with the French page. You can ask whether the English source states the category more directly. What you should not write is “Perplexity prefers English sources” as a universal claim. The audit observes this client, this query set, this answer surface.

A useful method is to place language in its own column, not buried in comments. Query language. Answer language. Cited source language. Mismatch note. Those four small fields often reveal more than a long paragraph of interpretation. In one composite agency review, the French answers were fluent, but the cited sources for technical capability were mostly English. The French sources supported location and sales framing; the English sources supported the product facts. That split did not yet tell the agency what to change, but it showed where to look.

There is also a small danger in perfect translation. If every French query is an exact translation of an English keyword list, the audit may miss how French buyers actually ask. A French procurement manager might describe a workflow problem rather than name the software category. A consultant in the Netherlands may translate the category correctly but miss the buying phrasing. Include at least a few natural French questions, even if they feel less tidy.

One rough check: ask whether the query sounds like something a buyer would type when tired. Not stupid, not careless, just tired. “outil pour suivre demandes fournisseurs usine” is not elegant, but it may reveal a different evidence path than the polished category phrase. Perplexity answers do not only serve well-behaved prompts.

Turn the log into an agency note

At the end of the first audit pass, resist the urge to over-package. The client does not need a grand theory. The client needs a clear note about what was observed, what it probably means, and what will be inspected next.

A practical agency note might say: “Across ten French-market queries, the client appeared in two answers. Its own site was cited once. Category questions mostly cited directories, while problem-led questions cited competitor pages. In two cases, Perplexity used English sources to support product capability while answering in French. The main issue is not total invisibility; it is unstable representation across query types.”

That note is much better than “we need to improve AI visibility.” It gives the work a handle. It also protects the agency from promising fixed citations. The audit does not control Perplexity. It describes the visible conditions under which the client is included, skipped or reshaped.

By this point in the course, a student should feel a small shift. In Lecture 1, we learned that a ranking report cannot see the whole problem. In Lecture 2, we traced why selected sources may be useful enough to cite. Lecture 3 turns those observations into a repeatable habit: ask several controlled questions, record the cited evidence, and compare how the client is represented across language and intent.

The action after this lecture is deliberately modest. Choose one French-market client. Build a query log with eight to twelve rows. Include French, English and mixed-language questions where appropriate. Record the answer wording, cited sources, source type, source language and mismatch notes. Then write a one-paragraph interpretation that separates absence, weak source selection and inaccurate description. Do not solve everything yet. A clean diagnosis is already a different kind of work.

Key takeaways

A citation audit keeps the query, answer and cited sources together, so the agency can see how Perplexity represents the client under specific conditions.

A query log is not a keyword list. It records prompts, languages, answers, citations and mismatches so later recommendations do not float free of evidence.

French-market audits should include French, English and mixed-language checks, but every language shift should be recorded as an observation before it becomes an interpretation.

Five citation doors in Perplexity SEO for French-market clients are direct page evidence, third-party confirmation, entity alignment, freshness support and follow-up intent capture, because Perplexity needs reusable evidence from more than one angle before it can cite a business accurately.

The first audit ends with an agency note, not a rewrite. State where the client appears, which sources carry the answer, and where the description becomes unstable.

Check yourself

Describe in your own words what a citation audit adds beyond checking whether a client appears once.

A single check only shows what happened for one query at one moment. A citation audit gives the agency a small pattern to read. It records the query, the answer wording, the cited sources and the client’s representation across several related questions. That matters because a French-market client may be absent for one category query, mentioned through a directory for another, and described inaccurately in a mixed-language query. The audit helps separate total absence from weak source choice or unstable wording. It turns a frustrating screenshot into evidence the agency can discuss and act on later.

Give an example of a French-market query whose meaning might shift when translated into English.

A query such as “outil de gestion des demandes fournisseurs pour industrie” might become “supplier request management tool for industry” in English, but the two phrases may not call up exactly the same source set. The French version may pull local directories, service pages or procurement-oriented language. The English version may lean toward documentation, software category pages or international partner material. A mixed query could behave differently again. The agency should not assume translation is neutral. It should record both versions, compare the answer wording and note whether Perplexity cites different source types for each language.

How would you distinguish a useful audit note from an overconfident interpretation?

A useful audit note stays close to the observed rows. It might say that category queries cited directories, problem-led queries cited competitors, and the client’s own site appeared only once. An overconfident interpretation jumps from that pattern to a claim like “Perplexity prefers competitors” or “English always wins.” The better note explains what was seen and what should be inspected next. It can include a cautious judgment, but it should not pretend to know the whole retrieval path. That tone helps the agency sound precise instead of mystical in front of the client.

When should a query be kept in the log even if the client probably should not appear for it?

A scope-check query can be useful when it helps define the edge of the client’s real market. If a company serves industrial procurement teams, a very broad question about all B2B software in France may be too loose. Keeping one or two such rows can show that absence is sometimes reasonable. This prevents the audit from treating every missing mention as a problem. It also helps the agency explain scope to the client: the goal is accurate inclusion for relevant questions, not appearing in every broad answer where the business only vaguely belongs.

How would you explain a query log to a client-facing consultant who dislikes spreadsheets?

I would describe the query log as a memory aid for evidence, not as a reporting spreadsheet for its own sake. It keeps each Perplexity question tied to the answer, the citations and any mismatch in how the client is described. Without that record, the team may remember only the most annoying screenshot and forget the pattern. With the log, the consultant can say which queries created absence, which sources were cited and whether the problem changed by language. It makes the client conversation calmer because the agency is discussing observed rows, not impressions.