LEYUN · SITE SPEC & COMPONENT LIBRARY独立站技术规范手册

FAQ

168 · 2026-11-20

Key takeawayThe most frequently asked technical and template questions across all sections. Each answer has a full article with comparison tables, diagnosis steps or code snippets behind it.

CRAWL & INDEX

What are the symptoms of an exhausted crawl budget?
Typical signs are new pages being indexed noticeably more slowly, updates to indexed pages taking a long time to appear, and crawl intervals for inner pages stretching out. In the GSC Pages report you will see many pages stuck in "Discovered - currently not indexed".
Will a site with only 500 pages run into budget problems?
Usually not. Very small sites almost never hit the budget ceiling. For small sites, the more common issue is crawl frequency being shaped by site size and response speed rather than budget allocation.
What is the difference between "Crawled - currently not indexed" and simply not indexed?
Both appear in the Pages report, but they mean different things. "Crawled - not indexed" means Googlebot visited the page and gave up on indexing it; plain "not indexed" usually means there is no crawl record at all, so the page was either never discovered or was blocked.
How good does content quality need to be?
There is no universal threshold. A practical test: among pages of the same type, identify what content and fields the indexed ones have that the non-indexed ones lack, then close the gap.
Does this state require active intervention?
Yes. It is the first break in the indexing chain; if the upstream fails, everything downstream is moot. But the fix depends on the cause: first diagnose whether it is a budget issue or a response-speed issue.
Does submitting a sitemap raise crawl priority?
A sitemap helps discovery and understanding of site structure, but it does not directly raise priority. Priority is mainly determined by internal linking structure and historical crawl performance.
Does Googlebot execute JavaScript?
It renders, but as a deferred process with limited resources and a rendering queue delay. The reliable approach in technical SEO is still to keep core content present in the HTML source.
Why do AI crawlers more often get blank pages?
Most answer engines' crawlers neither execute JavaScript nor perform deferred rendering; they read the raw response only. So the same page can show a snippet in Google results yet be blank in AI answers - a common phenomenon.
How long should logs be retained?
At least 30 days, ideally 90. Indexing problems often need to be traced back to when a change occurred, and that change may have happened weeks earlier.
Which log should be analyzed behind a CDN?
CDN edge logs are the most truthful. Origin logs miss requests answered directly by the CDN and therefore underestimate actual crawling.
Must all parameter pages be blocked?
No. The key is to avoid giving identical content multiple indexable addresses. Parameter pages with genuinely independent content can be kept, but the primary version must be declared with canonical.
Can canonical solve the parameter page problem?
It solves duplicate indexing but not budget consumption - the crawler still has to fetch each URL to see its canonical. Real budget savings come from robots blocking or server-side stripping.
Can paginated pages be indexed at all?
Yes, provided each carries independent content and a self-referencing canonical. The problem is that on most sites paginated content is repetitive, so the benefit rarely justifies the cost.
Is infinite scroll the better option?
It is friendlier to crawl budget but introduces new crawlability problems. It only beats pagination when paginated content is directly reachable by the crawler.
How long until a new site starts being indexed?
For sites with solid technical foundations, indexing typically begins within 2-4 weeks and stabilizes over 1-3 months. If it is still zero after 6 weeks, a specific technical cause definitely needs investigating.
Does unstable indexing matter?
Small fluctuations are normal, but large weekly swings usually mean the sitemap changes too often, pages conflict with each other, or unstable servers interrupt crawling.

RENDER & SCHEMA

Is infinite scroll always a bad idea?
You can use it, but pair it with server-side rendering and crawlable links. Products loaded purely by JS scrolling are essentially invisible to the index.
Does switching back to pagination hurt user experience?
Modern e-commerce widely uses pagination or "view more" buttons combined with infinite scroll. The key is having crawlable URL entry points, not just the visual scrolling effect.
What happens if hreflang and canonical conflict?
When the two signal sets contradict each other, search engines may ignore one of them or pick the most conservative interpretation. The safest setup is a self-referencing canonical on every URL that hreflang points to.
Does missing hreflang always hurt rankings?
Not rankings directly, but it does affect how international traffic is distributed and the accuracy of regional targeting.
Does every page need a canonical tag?
Strongly recommended, and indexable pages should always self-reference. A self-referencing canonical is the safest default.
What if the canonical does not match the URLs in the sitemap?
A sitemap lists URLs for discovery and crawl prioritization, while the canonical signals the chosen indexed version. When they disagree, follow the canonical and make the sitemap list only canonical URLs.
Is a bigger sitemap always better?
No. Prioritize URLs that have standalone content value and are not yet indexed. Stuffing the sitemap with low-value pages dilutes crawl attention.
Should redirecting URLs be submitted in a sitemap?
No. Redirecting URLs make parsers spend a request on a hop, and some parsers reduce their trust in the sitemap because of it.
Does breadcrumb markup affect rankings?
Not directly. Its main value is rich result display and click-through rate in search results, which indirectly affects clicks and downstream behavior signals.
Does every level need its own annotation?
Yes. The correct BreadcrumbList structure has one ListItem per level, nested via itemListElement to express hierarchy rather than listed side by side.
How similar does content have to be to count as duplicate?
There is no percentage threshold. The practical test: can search engines tell the intent of the two pages apart? If two pages give substantively different answers to the same query, they are not duplicates.
Should variant pages use canonical or noindex?
Variants with independent search value (e.g. separate product terms per color) get a self-referencing canonical and independent optimization; variants that exist only from parameter filtering get noindex.
What is the difference between returning 410 and 404?
404 means "not found for now"; 410 means "permanently removed." 410 makes crawlers stop retrying sooner and is the correct choice when deleting pages.
What should a 404 page contain?
A search box and the main section entry points, helping users and crawlers find other pages. The precondition is that the status code genuinely is 404.
Does a page need to be indexed by Google before AI engines cite it?
Not an absolute precondition, but the vast majority of cited pages are already indexed. Pages outside the index usually have no visibility on the AI side either.
AI crawlers fetch less than Googlebot — what does that indicate?
Usually that the content offers them too little value (duplicated, thin, or unextractable), not simply a crawl-budget issue.

PLATFORM & ARCHITECTURE

Is Shopify suitable for SEO?
Yes, provided you accept its fixed URL and rendering constraints and optimize proactively. Templates are highly standardized and competition is intense, so product-page differentiation matters especially much.
Does switching themes affect SEO?
It can. A theme change may alter URL structure, rendering, and template content. Record a URL comparison and indexing baseline before launch.
How do I quickly find which plugin is at fault?
Binary-search deactivation works best: disable everything to confirm the problem disappears, then restore in batches. Combined with curl TTFB tests using a crawler user agent, a few rounds usually pinpoint it.
What is the main SEO risk of WordPress?
Not the templates themselves but the uncontrolled output and performance fluctuation from the plugin chain. The more plugins, the higher the risk.
Do robots.txt-blocked pages still appear in search results?
Yes. robots.txt prevents crawling, not indexing. Already-indexed pages must be removed via noindex plus recrawling.
After a Disallow: / mistake, can the site recover?
Yes — fix the rules and request recrawling. But the index may have been progressively emptied in the meantime, and recovery takes time.
How much signal does a domain migration lose?
With single-hop 301s, complete mappings, and identical content, transfer is usually fairly stable. But ranking volatility in the first 2-3 months is common — monitor the recovery curve in GSC.
Should both domains stay live during migration?
Not recommended. The old domain should keep serving 301s but no content; both being simultaneously accessible produces duplicate content.
How soon after a redesign should indexing start recovering?
After technical fixes, recovery usually starts within 1-2 weeks; full recovery typically takes 4-8 weeks. If nothing moves in two weeks, an undiscovered technical problem remains.
Should the old site stay up during the redesign?
Yes. Keep the old site online with 301s until the new site's indexing is stable, then retire it gradually.
Subdomains vs. subdirectories — what is the actual SEO difference?
Mainly how fast crawl trust is established and deployment flexibility; the signal-transfer difference is far smaller than early myths suggested. Choose based on business independence, server arrangements, and certificate management cost.
How many subdomain sites can one domain support?
No hard limit, but each needs sufficient, independent content. The more thin subdomain sites you add, the worse the whole group performs.

PERFORMANCE

Do Core Web Vitals directly affect rankings?
Not as a direct ranking factor, but they are part of the page experience assessment and indirectly influence crawl frequency and user behavior signals.
What is the passing threshold for LCP?
Good is within 2.5 seconds, needs improvement is 2.5-4 seconds, poor is above 4 seconds. Judged on field data (75th percentile), not lab data.
How much latency is serious?
TTFB above 2 seconds clearly hurts both crawling and experience. 200-500ms is normal for cross-border access; above 1 second indicates a routing or configuration problem.
Can a CDN solve hosting latency?
Most of it. Once static assets and HTML are served from the edge, crawlers no longer cross borders to the origin — but the origin must still answer the first request correctly.
How do I verify a search engine's crawler IPs?
Every search engine publishes its crawler IP ranges in official documentation — use those lists and re-check them periodically. Keep dual verification: user agent matches AND the IP falls in the official range.
Could allowlisting crawlers be exploited for fake traffic?
Search engine crawlers generate no user visits and no direct traffic, so allowlisting them carries no fake-traffic risk — it is the precondition for indexing.
From which image should lazy-loading start?
The test is "outside the first screen." Above-the-fold images should not be lazy-loaded or LCP gets pushed back. A common pattern is prioritizing 1-2 first-screen images and deferring the rest.
WebP or AVIF — which to choose?
AVIF compresses better but encodes slower and has limited legacy browser support. The robust setup is AVIF + WebP + JPEG tiers delivered via <picture>.
Does HTTPS affect rankings?
Not as a direct bonus, but missing or misconfigured HTTPS causes crawl failures and reduced trust, indirectly affecting indexing and rankings.
Do subdomains need separate certificates?
Yes, unless you use a wildcard or SAN certificate covering all subdomains.
Which matters more: internal or external links?
For indexing and overall performance, internal links are the foundation: they decide how equity distributes and whether pages get discovered. External links provide initial trust and authority signals.
How many inbound links does a page need at minimum?
No hard standard, but key pages should have several inbound links from different sections. Inbound links all from one listing page have limited effect.
Do many broken links hurt rankings?
The direct effect is limited, but they indirectly affect indexing by consuming crawl budget and lowering link quality assessment.
How many 404s are serious?
Depends on share and source. If 404 requests exceed 5% of total requests, or concentrate on one path type, systematic handling is warranted.

B2B COMPONENT SPEC

How many columns should the spec table have?
Four columns is the most stable: parameter, value, unit, options. More columns get squeezed on mobile; fewer cannot express the options column, which matters greatly to buyers.
What if parameters differ between models?
Keep the header structure consistent and write "N/A" for non-applicable values. Columns jumping between models hurt readability and make side-by-side comparison by models harder.
Should I use a comparison table or a single-product spec table?
Both. The product page answers "what is this model"; the comparison page answers "which of these models should I pick". They match different stages of the purchasing process.
How many models per comparison table?
Three to six. More than six columns is nearly unreadable on mobile and exceeds the number of columns models can reliably align in a single extraction.
Should certificate images be included?
Yes, alongside text: text carries the number and validity date for search and extraction, images provide visual confirmation. With images only, the certificate number is essentially unsearchable.
Do certificates expire?
System certifications usually do; product certifications depend on the type. Any certificate with a validity period must show its expiry date and leave the main list once expired.
What should a factory capability module contain?
Verifiable objective facts: area, equipment count, capacity, process, quality control points. Not unverifiable adjectives.
Should headcount be shown?
Yes, but accurately. B2B buyers infer capacity and responsiveness from headcount; inflated numbers reduce credibility.
How many steps should the customization process have?
Four to six. Too few looks careless; too many makes buyers expect a long cycle. The common five: requirement confirmation, feasibility assessment, sampling, sample approval, mass production.
Should sampling cost be stated?
Yes, including whether it can be credited against a bulk order. Buyers weigh this when judging willingness to cooperate; vague handling lowers inquiry quality.
Which fields should a sample form collect?
Company name, contact person and work email are the baseline; target specification and usage description markedly improve sample matching. Never add fields you do not need.
Should sample costs be stated clearly?
Yes. Vague rules cause two problems at once: buyers with mismatched budgets never submit, and those who submit argue about the fee afterwards.
Must every case include numbers?
No, but any performance comparison must state its basis. Without data, a qualitative description is still more credible than "significant improvement".
Can case customers be anonymized?
Yes. Generalizing to industry and condition descriptions is common in B2B and lowers authorization risk.
Which matters more, FAQ or the product spec table?
The spec table says "what it is"; the FAQ says "can it do the job". Neither is optional, but the FAQ usually contributes more to conversion because it matches real purchasing doubts.
How many FAQ entries should there be?
5-12. Fewer than 5 usually leaves questions uncovered; more than 12 easily dilutes the topic.
Does a language switcher affect SEO?
The switcher itself does not, but misconfigured hreflang directly causes language versions to surface for each other's queries. It is the most common and most overlooked problem on multilingual sites.
Where should x-default point?
To the default-language version, or to a language selection page. The key is that the address must always be reachable.
Should downloads be gated?
Spec sheets can usually be downloaded directly; drawings and quote-related documents may require a work email. Excessive gating lowers both document retrieval and trust.
Should file names carry version numbers?
Not recommended. An update date matches engineering-document practice; version numbers make buyers misread them as different products.
How many required fields should an inquiry form have?
About four: company, contact, email, country or region. Everything else optional, and use dropdown options instead of text inputs wherever possible.
Is collecting phone numbers worth it?
Yes, but keep it optional. B2B buyers communicate by email; a mandatory phone field lowers both the submission rate and form completion.
How many breakpoints should a component library have?
One or two. More breakpoints raise the client's adjustment cost. One main breakpoint at 900px covers most scenarios.
Why insist that tables never hide columns?
Because mobile pages are crawled by search engines too. Hidden columns make key parameters invisible on mobile and directly hurt the page's content-completeness assessment.

MULTILINGUAL COPY SPEC

How long should a company profile be?
200-300 Chinese characters or 60-100 English words. Beyond that, buyers stop reading and model extraction slows down.
Should we claim "No. 1 in the industry"?
No. Unverifiable ranking claims are high risk in European and American markets and a negative signal to technical buyers.
How long should a product overview be?
40-70 words in English. Beyond that, scanning cost rises for buyers, and model extraction usually takes only the first paragraph anyway.
Should we write "high quality"?
No. Replace it with a verifiable statement such as "press force tolerance ±2% of rated value".
How long should the warranty period be?
12-24 months is common in export B2B. The key is a clear start point and coverage scope, not simply a longer duration.
How detailed should exclusions be?
Down to specific parts and cases (consumables, non-specified oil, unauthorized modification). Vague "normal wear and tear" is unenforceable in a dispute.
How long should a privacy policy be?
No fixed length, but it must cover: controller, collection scope, purposes and bases, retention periods, recipients, cross-border transfer, how to exercise rights, and contact. EU-market policies usually run 1200-2000 words.
Can all language versions share one policy?
EU member states implement privacy law differently. Localize for non-English markets, but the main structure can be shared.
How long is a typical preparation period?
7-15 working days for small-batch standard products, 15-30 for non-standard custom work. The key is stating the start point, not just the number of days.
Should freight amounts be written concretely?
Freight varies with weight and destination; usually write "calculated by destination and weight" and note whether it can be prepaid. Fixed figures quickly become inaccurate.
Must the responsible person be inside the EU?
Yes. The responsible party's address and contact details must be deliverable within the EU.
Can we still sell without an EU responsible person?
Most products bound by EU regulations cannot lawfully be placed on the EU market without one. Resolve this first.
How detailed should sample terms be?
At minimum cover: cost, crediting, lead time, freight, quantity limit, size difference, data conditions, appearance variance, usage restrictions. Nine items is the practical floor.
Should we write "samples match series production"?
Not as a blanket claim. Itemize what matches and what differs; it is accurate and lowers later dispute risk.
How long should the OEM page be?
200-400 words in English. The core is process, cost and intellectual property; beyond that, buyers stop reading item by item.
How should tooling ownership be written?
Three things: the owner, the effective condition, the exclusions. Writing only "transferred after full payment" leaves the generic-tooling dispute open.
How is one term maintained across four languages?
Use Chinese as the single key; each language version stores one value. Changing the term edits one place and takes effect on all four pages at once.
How many entries should the glossary have?
It depends on the parameter dimensions. Common industrial products land between 40 and 80 entries, covering the high-frequency parameters.
What does visible machine translation actually cost?
Mostly inquiry quality. Technical buyers who spot obvious machine translation judge the supplier unprofessional; dwell time and inquiry rate drop together.
What must never be machine translated?
Product parameters, certification and compliance statements, delivery terms, warranty terms, and anything containing numbers.

DATA TABLE SPEC

How many columns should a specification table have?
Four is the most stable: item, value, unit, remark. Reserve the remark column for operating conditions and variants; it is worth more than omitting it.
Must values always carry a unit?
Yes, but in a separate unit column. Mixed notation breaks parameter comparison and automatic charting alike.
How many models should a comparison table hold?
Three to six. More columns become unreadable on mobile and exceed what a model can align reliably in one extraction pass.
Does every model need its own detail page?
If the column headers link to detail pages, those pages must exist and match. Otherwise the table generates a large number of dead internal links.
How many rows should a material table have?
Cover the four to six materials you mainly offer. A longer table makes the material decision harder, not easier.
Should price ranges be included?
Not advisable. Prices move with the market and with order volume, so any figure on a long-lived page will go stale.
How many items should a tolerance table list?
Five to twelve critical dimensions. Listing every dimension leaves the reader without a focus and removes any guidance value.
Which standard should the general tolerance use?
Choose one per your drawing conventions and state the standard number. The essential sentence is "where not marked, this applies"; it is dispute prevention in itself.
How many entries should the certification list hold?
List the three to six main, verifiable ones. Padding the list with a dozen entries, several of them expired, lowers the credibility of the whole table.
What makes a verification path useful?
Write it as an executable step: which website, which menu, which number to enter. "Verifiable" on its own says nothing.
How many packaging types should be listed?
List the two to four you actually offer. Listing options you do not provide makes buyers doubt the rest of the page.
Are both net and gross weight required?
Yes. Freight is calculated on gross weight; a single figure only generates repeat questions.
How many bands should the table use?
Three or four. Too many bands force repeated cross-checking; too few leave most orders without a reference.
How often should the queue be updated?
As often as it actually moves. Daily in peak season, weekly in quiet periods. A stale queue is worse than none.
How do I confirm my table can be crawled?
Fetch the page source with curl and search for a parameter name. If it appears, it is in the HTML.
Is a more complex table better?
No. Merged cells and multi-level headers lower the parsing success rate; a simple, regular table is more robust.
Does every chart need a companion data table?
Recommended. The chart provides intuition, the table provides retrieval and citation; the cost is low and the role cannot be replaced.
How long should alt text be?
Long enough to include the key data points, usually 30 to 80 words. Shorter carries no information; much longer reads as machine-generated.

COMPLIANCE BANNER SPEC

Does the banner need three options?
At minimum accept and refuse; under most interpretations a category-level settings entry is also expected.
Should the banner appear immediately on page load?
It must not cover the main content. It may appear after load or after some scrolling, but not only after the user has visited several pages.
Does my product need this warning?
It depends on whether the product contains a listed chemical and whether exposure reaches the threshold. When unsure, consult a compliance adviser rather than assuming no.
Can the pop-up replace the product page warning?
No. The fixed product page warning is the baseline; the pop-up is an additional entry notice.
Which rights must be offered?
Access, correction, deletion, data portability and withdrawal of consent. Objection to automated decision-making also falls within scope in specific situations.
How far should identity verification go?
Proportionate to the sensitivity of the data involved. Verification that is too elaborate is treated as an obstacle.
Must I register in every EU country?
It depends on the product category and whether you sell there. Packaging covers almost all physical goods, but each member state runs a separate system.
Can registration numbers be published?
Yes, and publishing them is usually recommended; they are direct evidence of compliance.
Can a reviewer see information inside the dialog?
If the content is injected by JS, it is not in the page source. The safer approach is a static version on the page.
Should the dialog open automatically?
Not advisable. Compliance information should be findable on request; forcing a dialog interrupts browsing.
Is the age dialog the same as payment verification?
No. Age verification confirms age; payment and logistics carry their own compliance requirements.
Can a user still contact support after refusal?
Yes, and they should be able to. Some buyers click the wrong button and need a recovery route.
Is the Incoterms version really necessary?
Yes. Clause numbering and responsibility allocation differ between versions; only the version makes the applicable rules determinate.
Where should the notice sit?
One near the price on product pages, one in the cart, and a checkbox confirmation on the order confirmation page.
Do dialogs really conflict?
Yes. When a cookie banner and an age gate appear together, users do not know which to handle first. A priority queue that runs them in sequence solves it.
Is server-side region detection required?
When accuracy matters, yes. A server-side IP lookup written into an HTML attribute is the most reliable approach.

PDF PAGE SPEC

Does a download need a gate?
Datasheets are usually fine as direct downloads. A gate screens for real demand, but an overly strict one reduces both retrieval and trust.
Which comes first, the form or the direct link?
The form first (it captures more contact detail), the direct link after it as a fallback. Keep both.
Should the certificate list be its own page?
Yes. A separate page can be indexed on its own and is a natural landing point during procurement due diligence.
Will a scanned copy raise suspicion?
It can. Stating honestly that it is a reference copy and giving the official verification route is more credible than claiming it is an official electronic certificate.
What is the difference between an SDS and a TDS?
The SDS covers safety (composition, hazards, first aid, disposal); the TDS covers technical selection (performance data, applications, characteristics). Their purposes differ.
Does the SDS need translating?
Most target markets require a local-language version; English is the common minimum.
Can drawings be published openly?
Usually yes, but not where patented structures or customer-specific content are involved. Publish interface and overall dimensions and keep the internal structure out.
How many dimensions should the table hold?
Five to ten key dimensions. Full dimensioning belongs in the DWG file; the page only needs what affects selection.
Subdirectories or a flat structure?
Layer by type / model / language. With many models, a flat structure produces thousands of files in one directory and becomes unmanageable.
Must the URL file name match the real file name?
Recommended. Where they differ, downloads frequently fail with file-not-found errors.
Does PDF metadata affect ranking?
The effect is limited, but it affects being found and being cited. A PDF is an independent retrievable asset, and complete metadata means it can be classified correctly.
Which fields matter most?
Title and Subject. Author and Keywords play a supporting role.
Can PDFs be indexed by search engines?
Yes, depending on whether they are text-based, whether a landing page exists, and whether crawling is allowed. Missing any one of the three sharply reduces the effect.
How much text should a landing page carry?
A 300 to 600 word summary plus one key parameter table. That is enough for a search engine to judge the topic and for a buyer to judge fit.

GEO & AI RETRIEVAL

How does AI citation relate to classic rankings?
They are independent but share preconditions: the page must be indexed and the content must be accessible. Being indexed by classic search is one signal, but not a sufficient condition for AI citation.
Is dedicated "AI optimization" needed?
What is needed is shaping content into a parseable, extractable form with clear entities - which aligns closely with high-quality technical SEO and does not require a separate technology stack.
What is entity alignment?
It is the process by which a search engine maps the brands, products, and services described on a site to same-named entities already in its knowledge graph. Only after alignment can entity attributes be invoked correctly.
What should an entity description contain?
Verifiable, objective attributes: official name, founding year, location, category, official identifiers, and contact information. No subjective embellishment.
Can one site have both POI and Product entities?
Yes. A common brand-site combination is Organization for the company, LocalBusiness for stores, and Product for goods. The key is not to mix them.
Is LocalBusiness needed without a physical store?
Usually not. For online-only businesses, Organization plus Product is more accurate.
How can I tell whether an AI crawler actually fetched the page?
The most direct way is to send a request with that crawler's user agent and inspect the response body. Server access logs should also show these user agents' requests.
The AI crawler fetched the page but nothing was cited - what does that mean?
The problem lies in the markup or quality layer, not the technical layer. The most common causes are a lack of standalone extractable conclusions or inconsistent entity wording.
When is name ambiguity most severe?
When the brand name itself is a generic category word. The model then tends to read it as a general concept rather than a company entity.
Is renaming the brand the best solution?
Usually not necessary. Strengthening entity attribute identifiers, unifying site-wide wording, and building explicit Chinese-English correspondence already reduce confusion significantly.
Is daily updating mandatory?
No. A stable, predictable rhythm matters more. A high-quality site updating steadily every month usually beats a daily publisher with repetitive content.
What if updates are made but Google does not re-crawl?
Use GSC URL Inspection's "Request Indexing" to trigger a crawl, and confirm on the server that the sitemap lastmod has been updated.
Which matters more, mentions or links?
Links carry transferable signals; mentions establish the entity's presence in corpora. They complement each other: links build standing, mentions help entity recognition.
Do unlinked mentions have value?
Yes. They participate in the model's understanding of the entity and in how descriptions are generated. In Chinese-language contexts especially, a properly formed Chinese entity name matters more than large volumes of external links.
How do hreflang and canonical cooperate on multilingual sites?
Each language version self-references its canonical while hreflang declares the versions as correspondences of one another. The two have different jobs and do not conflict.
Should product model numbers be translated?
No. Model numbers are the most stable anchor for cross-language entity recognition; translation prevents the model from building the correspondence.
Does structured data increase the probability of being cited by AI?
It does not directly improve rankings, but it lowers the model's cost of recognizing page semantics, indirectly raising the probability of correct classification and citation.
Will FAQPage markup be displayed directly?
In some scenarios it is shown as a rich result. Display is decided by the search engine; the markup itself guarantees nothing.

Note: this page is a summary of publicly documented search-engine rules and is for self-check reference only. Engine behaviour is subject to the latest official Google documentation; verify against your own Search Console data.