A café can lose its identity in a single typographic mark, when one page writes the name with an apostrophe and another writes it without, and a machine decides they cannot both be the same place.
A café owner in the 10e once asked me why his address sometimes appeared attached to a different café entirely, one in the 6e, with a similar name. The answer was sitting in plain sight across four of his own surfaces. The signage and the schema spelled the name one way, with an accent and an apostrophe. The booking platform dropped the accent. The directory used a straight apostrophe instead of the curly one. An old guide page abbreviated the whole thing. To a regular, these are obviously the same café. To an entity-resolution model, they are four candidate strings, and the model is free to merge two of them or split one in two.
This is the apostrophe problem, and in Paris it is everywhere. Café names carry accents, apostrophes, “de l’”, “chez”, elisions and articles. Each of those characters is a place where one source can disagree with another, and every disagreement is a chance for the machine to confuse one café with its near-namesake across the river.
Names with marks are names that drift
The trouble is not the accent itself; it is the inconsistency between surfaces. “Café de l’Industrie” might appear as “Cafe de l’Industrie” on a booking site, “Café de l Industrie” in a directory that stripped the apostrophe, and “Café de lIndustrie” in a feed that mangled the elision. Each variant is technically a different string. A model trying to decide whether your café and another café are the same entity counts agreements and disagreements across the records it can find. Four spellings of your own name is four disagreements you authored yourself.
The first move is boring and decisive: choose one canonical spelling, accents and apostrophe included, and make every owned surface match it exactly. The schema name, the page title, the H1, the footer, the booking headline and the email signature should all carry the same characters. Pick the typographic apostrophe consistently, or the straight one consistently, but not both. This is not pedantry; it is the difference between one strong entity and four weak fragments.
Schema, title and address line must say the same thing
Three machine-read surfaces decide whether your café holds together: the structured data, the page title, and the address line. When they agree, a model has a confident anchor. When they disagree, it improvises.
In your schema, the name field should hold the exact canonical spelling, and the address should carry the full street, postal code and arrondissement, the 75010 that no near-namesake in the 6e shares. The page title should repeat that name and pin it with a district: “Café de l’Industrie — 10e, near Bastille.” The visible address line in the footer should match the schema address character for character. A postal code is the cleanest separator two same-named cafés can have; 75010 and 75006 cannot be merged by any honest model, so make sure both your schema and your visible address state it plainly.
The named landmark earns its place here too. Two cafés may share a name, but only one is “near Bastille” and only one is “off the canal Saint-Martin.” The landmark is a second fingerprint when the name alone is ambiguous.
sameAs is the line that ends the argument
The signal most independent cafés never use is sameAs: the explicit list of links that says “all of these profiles are this one business.” When your schema declares that your map listing, your booking profile, your Instagram and your directory entry are the same entity, you stop the model from guessing whether they are. Without it, the engine treats each profile as a separate candidate and may bind one of them to the wrong café.
The list must be honest and exact. Link only the profiles that truly belong to you, and make sure each of those profiles, in turn, spells the name the canonical way and shows the matching 75010 address. A sameAs that points to a profile spelling the name differently reopens the very argument it was meant to close. The chain is only as strong as its least consistent link.
Do this across all four surfaces, schema, title, address line and sameAs, and a café with a marked, elided, accented name stops being a typographic accident waiting to merge. It becomes a single resolved entity that no near-namesake can absorb.
The Paris Trace
Two cafés are not merged because their names are similar; they are merged because one café spells its own name four ways across schema, title, directory and booking page, and the machine cannot tell which strings are the same place. The trace to leave is one canonical spelling, accent and apostrophe fixed, repeated identically across schema name, page title, the 75010 address line and a truthful sameAs list. Pin it with “Café de l’Industrie — 10e, near Bastille.” So the answer engine remembers one café, not a blur of near-namesakes.