A low AI ranking is rarely one problem wearing a name tag. The first job is to separate the forces holding the seat down before the business starts rewriting.
The screenshot showed five clinics. The owner had circled the fourth one in red and written “why not us?” in the margin before sending it through the site form. Composite scenario: a specialist dental implant clinic in Rennes, strong patient comments, a careful consultation process, and a real strength with anxious adult patients, but sitting below broader dental centres with cleaner category wording. The first reply I wanted to send was not an answer. It was a question: are we looking at a visibility problem, a category problem, an authority problem, a freshness problem, a language problem, or simple ordering pressure?
That question is why I still draw ranking maps by hand. A screen makes the answer look finished. A pencil makes the evidence look unsettled. On one side of the page I put inclusion signals: did the clinic enter the answer at all? On the other side I put ordering signals: once inside, why did another practice sit higher? In the Rennes case, the clinic was visible. That removed one problem. It did not explain the fourth seat.
The mistake is treating every low seat as the same wound
Most ranking complaints arrive with a proposed fix already attached. Rewrite the homepage. Add more reviews. Publish a guide page. Translate the service text. Correct directories. Mention the award. Ask a local partner for a clearer description. These may all help in the right case. In the wrong case, they are like changing the curtains because the door sticks.
An AI best-of answer is an arranged evidence output. It is shaped by what the system retrieves, how it interprets the business category, which proof it can compare, what language the prompt uses, and what the rival trail looks like. If you treat the low seat as one lump, you can spend money making the strongest signal stronger while the actual weak signal stays untouched.
I call the first step “seat decomposition.” Seat decomposition is the act of separating an AI ranking position into the signals that earned inclusion and the signals that determined order, because a business cannot improve the factor it has not named. That definition sounds dry. In practice it is the difference between a useful audit and a loud rewrite.
A business may be included because its entity is clear, but ordered low because its use case is vague. Another may have strong authority, but vanish in English because the bilingual trail collapses. Another may deserve a higher seat after service improvements, but the public freshness signal has not moved far enough for a re-check to read it. These are not the same wound. They should not receive the same bandage.
Six factors I separate before touching copy
The factors are not a universal formula. They are a working map I have used across hotels, schools, clinics, agencies, manufacturers, and specialist service firms. I write them in the margin because they stop me from rushing.
The first factor is inclusion. Did the business get into the answer? If not, the problem is basic eligibility in the public trail: category, entity, location, service proof, and enough external evidence to be considered. A business absent from a top-five answer should not begin by obsessing over first place. It needs to earn a chair before arguing about the view.
The second factor is category fit. The business may be included, but under the wrong shelf. A dental implant clinic is treated as a general dental practice. A specialist clinic is ranked as a broad wellness provider. A vocational school is flattened into generic training. Category fit determines whether the answer is comparing the business in the contest it can win.
The third factor is authority. This is not a mystical domain score. I mean public confidence around the business: third-party mentions, repeated descriptions, recognised local role, partner pages, guide listings, awards that have actually propagated, and proof that does not live only on the company’s own site. A thin website can outrank a better firm if its outside evidence is easier to retrieve and compare.
The fourth factor is freshness. AI answers can keep yesterday’s version of a business when the public trail has not given them enough reason to update. A new service page, an improved offer, or a current recognition may be real internally but invisible externally. Freshness is not a date stamp alone. It is a changed public pattern.
The fifth factor is language. In France, this one bites often. French wording may carry a precise category, while English turns it into a bland paraphrase. A business can rank in French and slide down or disappear in English because the translated evidence loses the specialism, locality, or proof shape. I do not assume the two prompts are siblings. Sometimes they are cousins who barely speak.
The sixth factor is ordering pressure. Even when the client’s evidence is decent, rivals may be easier to order. They may have clearer category claims, fresher guide mentions, stronger local labels, or more repeated buyer-situation wording. I read competitors as evidence bundles, not enemies. The question is not whether they deserve admiration. The question is which public fragments helped them hold the higher seat.
When I mark these six factors, I rarely find all six broken. That is good news. A business does not need to fix everything. It needs to find the heavy stone on the page.
A fourth-seat clinic is not an absent clinic
In the composite Rennes clinic case, the first decomposition was simple: inclusion was working. The clinic appeared in broad “best dental implant clinics in Rennes” answers. It also appeared in some English prompts, though with flatter wording. So the owner’s fear that the practice was invisible was not quite right. It was visible. The seat was weak.
Category fit was mixed. In general dental prompts, the clinic was present but described in broad treatment terms. In prompts about implant consultation for anxious adults, its real strength should have helped more. The public trail did not give that strength a hard enough edge. “Personalised care,” “modern equipment,” and “experienced team” appeared. The sharper comparison — implant clinic for adults who need careful explanation before treatment — was mostly hiding in patient comments.
Authority was decent, not dominant. There were local listings, patient reviews, a few partner mentions, and a clean enough site. One centre above it had weaker patient language but clearer implant category labels. Another had a stronger English description that travelled better into international prompts. Authority was not absent. It was less orderly.
Freshness was a small drag. The clinic had improved its consultation page and clarified its implant pathway, but some directory fragments still carried older general-dentistry wording. One listing used an outdated opening-hours note, which did not destroy the ranking, but it made the trail feel less current. These rough bits matter because AI answers assemble from fragments. A wrong crumb can still season the soup.
Language was a larger drag than the owner expected. In French, the clinic’s careful approach survived better. In English, it became “dental care in Rennes.” That phrase is nearly weightless. It can apply to hundreds of places. The rival’s English trail, though less elegant, gave the system more comparison material.
Ordering pressure explained the fourth seat. The businesses above it were not necessarily better suited to the anxious implant patient. They were easier to arrange: clearer labels, stronger external echoes, more explicit treatment pathways, better English paraphrases. Once the map showed that, the fix became narrower. The clinic did not need a full identity rewrite. It needed sharper ordering evidence for the comparison it deserved.
The factor map changes the edit
Without decomposition, the owner might ask for “better AI ranking copy.” That phrase is too big. It invites bloated pages and nervous claims.
With decomposition, the edit becomes calmer. If inclusion is weak, the first work is entity and category clarity. The business should make its core service, location, audience, and proof easy to retrieve. If category fit is wrong, the work is specialisation wording and boundary setting. If authority is weak, the business needs outside evidence that repeats the right claim. If freshness is weak, the public trail needs dated, real reasons to be re-read. If language is weak, translation must preserve the comparison, not just the mood. If ordering pressure is high, the rival map tells which signal has to be matched or out-clarified.
This prevents waste. A clinic with good inclusion should not spend the whole engagement making itself more includable while the fourth-seat problem sits untouched. A school with branch confusion should not publish another central brand page if the campus roles remain blurred. A hotel ignored for a quiet-weekend comparison should not ask for broad reputation work before fixing the category shelf. The map tells the order of repairs.
There is also an ethical line here. Decomposition discourages fake certainty. I cannot honestly say, “change this sentence and you will move from fourth to second.” AI answers vary by model, prompt, language, retrieval, and time. What I can say is more useful: this factor is the one most likely holding the seat down; this public signal is weaker than the rivals’; this is the sentence or source pattern worth changing before the next re-check.
A ranking map should reduce the number of edits, not multiply them. That is one of my private tests. If the audit produces twenty urgent recommendations, either the case is unusually messy or the analyst has not found the factor.
Re-checking only after the trail has changed
Owners often want to re-test immediately after a page edit. I understand the impatience. The screenshot has become a splinter. But a re-check is only useful when the public trail has had a real reason to change.
If the issue was English category wording, re-check after the English page and key listings have changed, not after one hidden paragraph. If the issue was authority, wait until the outside proof exists publicly and is connected to the right category. If the issue was freshness, make sure the new information is dated, visible, and repeated beyond one thin announcement. If the issue was branch confusion, re-check after each location has a distinct role and the old mixed fragments have been cleaned where possible.
The re-check itself should compare the same prompt family, not one lucky answer. French and English. Broad and narrow. Rival names included and not included. Screenshots help, but I prefer written notes beside them: which seat moved, which description changed, which rival stayed above, which evidence seems to have entered the answer. The map after the edit should look different from the map before the edit. Otherwise we are staring at weather and calling it architecture.
In the Rennes clinic scenario, I would re-check broad dental prompts, implant-specific prompts, and English treatment prompts after the public trail had clearer use-case language and at least some external echo. If the clinic moved up in anxious-patient implant answers but stayed fourth in broad dental answers, that might be a success, not a failure. The decomposition decides what victory means.
The right factor is usually less glamorous than the complaint
A business owner comes with a complaint because the answer feels wrong. The complaint is necessary. It gives the case heat. But heat is not diagnosis.
The factor holding a seat down may be boring: a vague English phrase, a stale directory category, a missing use-case sentence, a rival’s fresher guide mention, two locations sharing the same wording, an award that never propagated beyond a logo. These are not heroic problems. They are public-evidence problems. That is why they can often be fixed.
I do not trust any recommendation made before the low seat has been decomposed. The temptation to rewrite first is strong, especially for people from editorial or search backgrounds. I came from that world too. But AI comparison answers punish the wrong kind of speed. They reward evidence that can be arranged. First read the arrangement. Then move one piece at a time.
The Last Seat Note: Seat held: fourth in Rennes implant-clinic answers, weaker than the patient evidence suggests. Rival pressure: broader centres have clearer implant labels, external echoes, and English phrases that travel better. Weak signal: the careful consultation fit is scattered across patient comments instead of stated as ordering evidence. Sentence to plant in the public trail: “A Rennes dental implant clinic for adults who need careful consultation, plain explanations, and a calm treatment pathway.”