What have authoritative sites measured recently?

On 2026-09-08, Semrush published “AI search & manufacturing SEO: What the data shows.” The choice of topic is worth noting in itself — they did not lump manufacturing in with general B2B, but treated it as an industry with its own patterns to test.

All I have is the title and the direction, not their raw data tables, so I cite no specific percentages here. What I adopt is the question they raised: search behavior, buying paths, and information needs in manufacturing differ from consumer goods, and forcing them into the same keyword-selection and content templates distorts the result. That reminder matters especially for export factories, because most SEO tutorials are written for e-commerce and SaaS.

Put this together with two other public data points and the direction becomes clearer. On 2026-07-02, Semrush measured AI Overviews expanding into commercial-intent queries; in a study of 15 million queries dated 2026-10-12, Ahrefs showed that a single prompt gets broken into multiple sub-queries. For manufacturing, that means when a buyer asks one question — “what material should be used under these operating conditions” — the system behind it may be searching along several lines at once: material, process, service life, cost. Each of those lines can be an entry point where you get cited.

How is AI search in manufacturing different?

First, the questions are longer and narrower. A consumer asks “which one is better”; a manufacturer asks “304 or 316 in a seawater environment — which is more suitable, and how much does the cost differ?” Longer questions demand more from content: you have to cover material, operating conditions, service life, and budget at the same time, and state where each answer applies and where it does not.

Second, the decision chain is longer. A buyer may first ask an AI, then check Reddit or an industry forum for real feedback, and only then contact a supplier. According to Ahrefs research dated 2026-07-27, Reddit appears in 83.9% of Google’s discussions and forums results, which shows how much weight “real experience” carries in search. And technical discussion in manufacturing is exactly what is most active on forums.

Third, directory sites occupy a lot of space. Search a broad category term and what ranks at the top is often B2B platforms and industry directories, not a factory’s own site. Winning at that level is hard for a factory, but switch to specific operating-condition terms and directory sites usually have nothing to match, and that is the gap. For the approach to choosing keywords, see competitor and keyword research.

We tore down 12 competitors. What did we see?

I ran a round of competitor teardown, picked 12 peers, extracted 53 non-brand keywords, and looked at the composition of the top ten results one by one. I expected to find a batch of keywords where “no strong site ranks in the top ten.” The count was 0. They are either held by platforms and directories or guarded long-term by two or three established sites. The full record is in field research and methodology notes.

That result sounds pessimistic, but it points the direction out clearly: there is no winning a head-on fight for broad terms. The real gaps are in specific long-tail questions, especially ones that involve operating conditions, failures, or comparisons. These keywords have low search volume and a messy top ten, which is precisely what gives factory content its chance.

Another usable signal comes from our AIO test of 500 Chinese-language queries: technical queries triggered an AIO 62.9% of the time, clearly below the 96.7% for tutorial terms. This suggests technical questions have not been fully absorbed by AI summaries, and users still need to click through to the source to verify. For a factory producing technical content, that is a window that has not closed yet. Plan it alongside how to gain visibility for product and solution queries.

Which kinds of keywords should a factory choose?

My ranking puts operating-condition terms and failure terms first. The reason is practical: only people who have actually done the work can answer these two kinds of questions, AI cannot invent the details, and directory sites have even less incentive to write them. Write an article on “three rounds of troubleshooting porosity in a certain type of casting” and you have content that is unique by nature.

Process terms come second, because they carry parameters and lend themselves to comparison tables. Broad category terms come last: one page that covers the category completely, doing the job of being navigated and linked to, is enough. For the overall framework, see AI search visibility for B2B exporters.

How should pages divide the work?

One page answers one question

Split “how to troubleshoot porosity” and “how to prevent porosity” into two pages; do not merge them into one long article. When AI extracts an answer it prefers pages with clear boundaries, and readers find the passage they need more easily too. I broke down how to structure them in page structure and citation probability.

Product pages carry parameter tables; blog posts carry judgment

Product pages handle specifications and inquiries, so write out parameters, materials, tolerances, and lead times in full. Blog posts handle judgment, such as how to choose between two processes or where the difference in quotes comes from. Link the two to each other; do not let a product page turn into a technical paper, and do not let a blog post take inquiries.

Leave one piece of verifiable evidence on every page

Photos, inspection reports, equipment lists, delivery records — pick any one. Manufacturing buyers trust images and data far more than copy. English sites need particular care here: do not write only in Chinese and then machine-translate. For how to write, see how an export English site gets cited by AI.

A low-cost action checklist

Three steps, no new tools, and the first batch of content can be out within two weeks.

1. Collect 20 real questions from the shop floor: have sales and engineering each list ten questions customers have asked. Write down their exact words; do not polish them.
2. Rank by operating condition and failure first: pick ten that only you can answer, and write the first three.
3. Give every article one on-site photo and one number: the photo must be one you took yourself, the number must be real, and do not use stock libraries.

Once this round is done, you will have three pages of content no one else can copy. To systematically monitor whether AI cites them, see the AI visibility monitoring checklist.

The questions owners ask most often

How is doing AI search for manufacturing different from ordinary SEO?

Demand is more concentrated on specific processes, parameters, and operating conditions, and the questions are longer and more technical. Applying the general SEO approach to keyword selection tends to land on terms with no purchasing intent.

Our keywords are all very technical and have low search volume. Are they worth doing?

Yes. Technical keywords have low volume, but each search sits close to a purchasing decision. Ten low-volume, high-intent keywords are usually worth more than one broad term.

Why is it always the same few big brands in AI answers?

Big brands have more public information and a consistent message, so models cite them more easily. The opening is not head-on competition but the specific operating conditions and parameter boundaries they fail to spell out.

We do not have a technical team. How do we write the parameters clearly?

Have the people in the workshop and engineering talk it through out loud, and have marketing turn it into Q&A. One article answering one question, with the conditions where it applies attached, is more effective than writing a complete product manual.