A corrected Google Business Profile can still lose the argument if the wider web tells an older story. AI answers read Bangkok through conflict, not through one dashboard.
A recurring composite café case starts with a Bangkok owner updating Google Business Profile hours before a holiday period. The map card looks right. The owner relaxes. Then an English AI answer tells a visitor the old closing time, using the phrasing from an old food listing. The rough detail is familiar: the Thai Facebook post has the current hours, the profile has the current hours, but the English listing still has the sentence a visitor can quote. The visitor arrives annoyed, phone in hand, as if the business had personally promised the wrong hour.
This is a familiar kind of modern irritation: the owner fixed the source they could control, but the answer came from a source field they could not see. In Bangkok, that field is crowded. Map facts, hotel pages, mall pages, delivery listings, Wongnai-style food pages, travel sites, old blog posts, booking platforms, Facebook snippets, and copied directory text all sit around the same business. AI does not always treat the newest profile edit as the final word.
The profile is strong, but it is not the whole city record
Google Business Profile is one of the strongest public identity surfaces for a local business. It carries address, hours, category, photos, reviews, phone, branch labels, and map position. For many Bangkok operators, it is the most actively maintained source they have. But AI answers are assembled from more than one public surface, and the conflict between surfaces matters.
The mistake is assuming that one corrected profile automatically cleans the answer layer. It may, over time, influence the visible map environment. It may also help answer systems that consult map-like sources. Yet many generated answers still lean on old pages because those pages contain sentences that are easier to quote. A profile says “Open until 22:00.” An old directory says “This Ari café is open late for dinner and drinks.” The directory sentence may be stale, but it is narratable.
I call this profile-source split. Profile-source split is the disagreement between a business’s controlled map profile and the broader public text AI uses to describe that business. The split shows up around hours, address, branch identity, category, menu, prices, venue relationship, and audience. It is especially common in Bangkok because businesses are often described differently for Thai locals, hotel guests, tourists, expats, and delivery customers.
The fix is rarely to edit the profile again and again. The fix is to find which outside sentence still carries the wrong fact.
A hospitality composite with three public stories
A typical composite from the hospitality side has one long-running Thai restaurant, one mall branch, and a rooftop bar attached to a boutique hotel. The operator maintains its profiles reasonably well. The original restaurant’s map card has updated hours. The mall branch has a separate pin. The rooftop bar is named on the hotel site and on some booking surfaces. To a human who knows the group, the structure is clear enough.
AI answers still disagree.
One answer pulls old hours from a food listing for the original restaurant. Another describes the mall branch as if it were the original venue, because a directory uses the same English name without a branch handle. A third folds the rooftop bar into the hotel and treats it as an amenity, not a separately chosen venue. The Google Business Profiles are not useless here. They are part of the evidence. They just do not settle every public contradiction.
The imperfect detail matters: one AI answer correctly named the mall but used the wrong phone number. That kind of mixed accuracy makes the owner’s job harder. It proves the system is seeing some current information while still carrying an old piece from elsewhere. The problem is not ignorance. It is unresolved conflict.
Bangkok branch businesses suffer most from this because one name can live across different floors, malls, hotels, neighbourhoods, delivery zones, and Thai-English spellings. A single old listing can leak across the whole structure if the official branch pages do not repeat the distinction in plain text.
Which signals usually settle the conflict
I do not treat all public signals as equal. In field notes, I rank them by how much they seem to help an answer resolve a specific conflict. For hours, the strongest repair is usually consistency across the profile, official page, and any high-visibility directory that still appears in answers. For address, the repair needs the map profile and an official location sentence that names the city handle. For category, the official page must explain the business type more clearly than the directory label. For branch identity, each branch needs its own page or section with the branch name, area, phone, hours, and service difference.
A profile edit without an official sentence can be too quiet. An official sentence without a profile correction can be contradicted by the map environment. A directory correction without branch structure may clean one surface while leaving the model confused about entity relationships.
This is why I use a conflict ledger for local AI answers. The ledger is simple: answer claim, business-controlled source, third-party source, conflict type, likely stronger signal, repair sentence. It is not glamorous work. It is closer to checking a shop-house ledger in pencil than running a grand visibility campaign. But the pencil work catches the exact point where the answer went wrong.
A Bangkok business should treat its Google Business Profile as the coordinate source and its official site as the explanation source. AI answers often need both: one to locate the entity, the other to describe it.
Hours, categories, and branches fail differently
Wrong hours are the easiest for owners to notice. They create immediate friction. A customer arrives when the kitchen is closed, or expects a rooftop bar to serve food at a time when only drinks are available, or repeats a holiday schedule that belonged to a past period. The repair is freshness language. Not a vague “check our page for updates,” but a visible current-hours sentence near the relevant venue or branch page. If hours vary by branch, write them by branch. If holiday hours change, keep a durable note that says where official updates appear.
Wrong categories are quieter. A restaurant becomes a café. A clinic becomes a spa. A school becomes a language service. A hotel restaurant becomes a hotel facility. Category errors often come from directories, because directories force a business into a fixed label. The official site needs one clean sentence that names the category and the boundary. “This is a separately named rooftop bar inside the hotel” is stronger than three paragraphs of atmosphere.
Branch errors are the most tangled. Bangkok businesses often use one brand name across branches while customers use mall names, BTS habits, district labels, or Thai nicknames. If the map profile says one thing, the delivery listing says another, and the official site groups every branch on one thin page, AI may mix phone numbers, hours, menus, and reviews. The repair is entity separation: each branch must be written as a distinct answerable unit.
The old source may still be useful
Owners often want to remove or ignore old listings. Sometimes that is reasonable. But in many cases, the old source is still part of the city record. A travel platform may still rank. A food directory may still be quoted. A hotel page may still describe the venue. A map snippet may still carry the branch name people use. The question is not whether the business likes the source. The question is whether AI appears to use it.
I do not recommend chasing every weak page on the web. That becomes noise. I look for sources that repeat inside answers or seem to provide the wrong phrase. If an AI answer says “near Silom nightlife” and that exact wording appears on an old travel listing, that source deserves attention. If an answer invents a menu item and an old directory page contains a similar dish name from a past branch, that is more useful than ten generic mentions.
The repair can be direct correction, stronger official wording, a branch page, a venue relationship line, or a public note that clarifies old versus current facts. The right move depends on the conflict. A corrected Google Business Profile starts the process. It does not finish it.
For profile conflicts, the contact form can start with the answer and the public sources that disagree. The useful question is not “why is AI wrong?” but “which source taught it that particular wrong fact?”