A heading is not just a label above a block of text. In an AI answer, it may become the handle by which the whole page is picked up, moved and misread.
The heading looked elegant in the browser. “At your side, wherever production cannot stop.” A nice line. Too nice, perhaps. Below it sat three paragraphs about maintenance, continuity and regional presence. A human reader from the sector could probably work out the offer. The answer engine did not. It called the company a “general repair provider,” which is almost true in the same way a bread knife is almost a saw.
The composite scenario is a 12-person industrial maintenance company in Auvergne-Rhône-Alpes. It services food-processing equipment across several departments. The Google Business Profile is decent, the reviews are strong, and the team has real field knowledge. But the page headings are all mood and no machinery. “Fast support for your operations.” “A local team that understands your needs.” “Maintenance with confidence.” When I compare those lines with a competitor’s blunt heading — “Preventive maintenance for food-processing conveyors in Loire and Haute-Loire” — the reason for the AI answer becomes less mysterious.
Headings are load-bearing text
In classic SEO work, we learned to treat headings as structure and signal. The H1 named the page theme. H2s divided sections. Keywords appeared there because search systems and human readers both used them to understand the page. That habit still matters, but answer engines add a harsher test: can the heading travel outside the page without needing repair?
A liftable heading is a heading that can stand as a factual claim, because it names the service, entity or boundary clearly enough to survive extraction. It does not need to be ugly. It does need to carry weight.
Many French SMB pages use headings as mood lighting. They set tone, reassure the buyer, create a smooth path through the page. “A partner for your performance.” “Expertise close to your teams.” “Solutions adapted to your constraints.” These are not crimes. They are just weak handles. If an answer engine is trying to decide whether the page supports a specific service recommendation, those headings give it very little to hold.
The heading is often the first sentence-like object a model sees around a block. If the heading is vague and the paragraph below is also soft, the whole section becomes fog. If the heading is precise, the paragraph can explain, qualify and persuade. Precision at the top changes the reading of everything below it.
A clear heading gives the model a shelf label and a source claim at the same time. That is why a blunter competitor sometimes wins the citation even with a less attractive page. The competitor has written something a machine can quote without surgery.
The problem with elegant French headings
French service pages often have a particular kind of polished vagueness. I notice it because I have written some of it myself. The language tries to sound serious without being heavy, local without being provincial, expert without being dry. It is easy to end up with headings that feel balanced in the meeting and collapse in the extraction.
“Your industrial maintenance partner in the region” looks acceptable. What region? Which industry? Preventive or corrective? Machinery, buildings, electrical systems, production lines? Does the company serve factories, restaurants, logistics sites, farms? The page may answer later, but the heading does not.
In the maintenance-company composite, the most important service boundary was food-processing equipment. The most important place boundary was several departments in Auvergne-Rhône-Alpes, not the whole region and not just one city. The most important action was preventive and corrective maintenance, especially on production equipment that could not sit idle. None of that appeared in the main headings.
The answer engine did what answer engines often do with soft copy. It generalized. It read maintenance as repair. It read regional language as local service. It read food-processing equipment as industrial equipment. Each step seemed small. Together, they changed the commercial meaning of the business.
One small roughness made the case more annoying. In one run, the model named the company correctly and even placed it in the right broad region, but then described it as serving “commercial kitchen equipment.” That phrase did not come from the company page. It looked like a borrowed category from somewhere else. The page had left a gap, and another source had poured language into it.
I use the three-load test
When I audit headings, I use a simple classification I call the three-load test. A heading can carry topic load, service load or proof load. Strong pages usually need all three across the page, though not in every heading.
Topic load tells the reader what category they are in: “Industrial maintenance in Auvergne-Rhône-Alpes.” Useful, but broad.
Service load tells the reader what the company actually does: “Preventive and corrective maintenance for food-processing equipment.” Better.
Proof load gives the claim a concrete surface: “Maintenance visits documented with intervention reports for production teams.” Now the section is not just announcing a service. It is showing how the service leaves evidence.
Weak pages lean almost entirely on topic load or mood. They say “maintenance,” “support,” “expertise,” “proximity,” and hope the paragraphs fill the rest. Stronger pages distribute the load. One heading names the service, another names the operating boundary, another introduces evidence.
A heading that carries only mood may help the page feel pleasant, but it rarely helps an answer engine identify the service. Mood can remain in the body. The heading should do the harder work.
For the maintenance company, I would not rewrite every heading into a dull label. That would make the page sound like a parts catalogue. I would change the load-bearing headings first.
“Support for your production challenges” could become “Maintenance for food-processing production equipment across Loire, Rhône and nearby departments.” That is longer, and perhaps a little square. Still usable. It tells the truth.
“A responsive team near you” could become “On-site interventions for recurring equipment faults and planned maintenance visits.” The word “responsive” may remain in the paragraph, where it can be supported by response rules, scheduling notes or examples.
“Know-how that protects your activity” could become “Intervention reports that document faults, repairs and maintenance recommendations.” That heading carries proof load. It gives the model and the buyer something firmer than confidence.
H1s fail when they try to be slogans
The H1 is the worst place to be coy. It is often the strongest single page signal, and many businesses spend it on a phrase that could belong to anyone.
“Your partner for industrial performance” is not a bad slogan. It is a weak H1. It does not identify the business with enough force. I would rather see an H1 that says, “Industrial maintenance for food-processing equipment in Auvergne-Rhône-Alpes.” It is not poetic, but it lands.
There is room for a supporting line below it: “Preventive visits, fault diagnosis and documented interventions for production teams that cannot leave equipment idle.” That line adds action and context. The pair gives the page a stable extraction unit.
This is where I sometimes argue with designers. They want a short H1 because the layout breathes better. They are not wrong. A page is visual. But if the short line removes the service boundary, the page becomes beautiful at the cost of being misdescribed.
There is a compromise. Use a precise H1 and make the design carry the elegance. Or use a short H1 only when the subheading immediately completes the claim. What fails is the combination of a vague H1 and a vague subhero. Two soft lines do not make one clear one.
An AI answer is unlikely to preserve your visual hierarchy; it preserves the statement it can understand. That is a brutal little rule, but useful.
The H1 should answer the basic extraction question before the page asks for trust. Who is this for? What is done? Where or under which boundary? If those facts are missing from the top, the page makes the rest of the internet speak on its behalf.
How to keep headings human
A page full of exact factual headings can become tiring. I do not want every French business site to sound as if it were written by a procurement department. Headings need rhythm. They help a reader move. Some can be short. Some can be warm. The issue is sequence and responsibility.
I usually choose two or three headings on a page to become factual anchors. These are the headings near the core service, the evidence section and the location or operating boundary. Other headings can be lighter, as long as they do not replace the anchors.
For example, this set is weak:
“Maintenance that moves with you.” “Our expertise.” “A trusted local presence.” “Your equipment, our priority.”
This set is stronger without becoming inhuman:
“Preventive maintenance for food-processing equipment.” “Fault diagnosis and documented repair visits.” “Service coverage across selected departments in Auvergne-Rhône-Alpes.” “Keeping production interruptions legible before they become expensive.”
The last heading has a little more voice. It still points to a real mechanism: interruptions need to be documented, understood and prevented. The page does not have to choose between machine clarity and human tone. It has to stop making every heading carry only tone.
One teaching example makes the difference clear. Imagine an answer engine comparing two pages for the query “industrial maintenance food-processing equipment Auvergne.” One page says “Your local performance partner.” The other says “Maintenance for food-processing conveyors and packing equipment in Auvergne-Rhône-Alpes.” Even if both companies do similar work, the second page has given the model a safer sentence.
The first company may still be better. The page has failed to say so.
The heading should reduce repair work
I often think of AI extraction as repair work. The model receives fragments from pages, profiles, directories and summaries. Some fragments are already clean. Others need patching. A vague heading forces the model to repair the sentence before using it. Repair introduces error.
A good heading reduces repair work. It does not ask the model to add the service, infer the place, guess the equipment or soften the claim. It gives a compact version of what the business can stand behind.
This matters because headings are copied, skimmed, embedded in snippets, placed near schema, and repeated in internal links. They become small pieces of the site’s evidence pattern. If those pieces are vague, the pattern is vague even when the business itself is not.
The maintenance company in the composite did not need a new identity. It needed headings that behaved like labels on drawers in a workshop. Not decorative labels. Real ones. The drawer marked “bearings” should not say “movement excellence.” Someone with greasy hands needs the part.
So does the model.
The Lift Note
Query: “titres H1 pour IA.” Liftable sentence: “A liftable heading names the service, boundary or proof clearly enough to survive extraction outside the page.” Missing proof: headings that carry service load, place load and proof load instead of mood alone. Rewrite instruction: make the H1 state the core service and market boundary, then revise two H2s into factual anchors for service detail and evidence.