TL;DR
- A mention, a citation of your own page and a qualified inquiry are three different results. Report them separately.
- Sector data shows mentions, citations and AI referrals do not move together. No single one shows buying interest.
- Buyers research alone but still want expert input on fit. Publish approved capability proof instead of broad claims.
- Use a capability proof matrix to decide what each page must show, who signs it off and where fit review starts.
Three results that are easy to confuse
Industrial purchases turn on fit. Can this supplier machine our alloy, hold our features, document the inspection and deliver at our volume? A brand name in an AI answer settles none of that.
Keep three results apart. A mention is your name in the answer text. An owned citation is a link to one of your own pages as a source. A qualified inquiry is a request that meets fit rules you set in advance. Our measurement guide sets out counting rules for the first two.
For planning, review the facts behind brand mentions, the evidence on cited pages and the route from those pages to an engineer. Each review answers a different question.
What the manufacturing data shows
In Semrush’s manufacturing study (September 2026), mentions, citations and AI referrals told three different stories. It combined three datasets for January to July 2026, including answers from ChatGPT, Gemini, AI Mode and AI Overviews.
Its lists diverged. Only 2 of the top 15 most-mentioned domains were also among the 15 most-cited. Product brands led mentions; directories, distributors and reference sites featured among citations.
Referrals were small. AI assistants and AI Mode sent 0.48% of US sessions to sites in ten manufacturing categories.
One pool of clickstream sessions. Smaller channels are not shown. Semrush manufacturing study, September 2026.
What the data does and does not tell a manufacturer
A small referral share does not mean AI answers have no influence. A buyer may read an answer, then search your name or type your address days later. Analytics would log that as branded search or direct traffic. The reverse also holds: rising direct traffic does not prove AI caused it. Treat that gap as unmeasured, not as hidden success.
For planning, the citation pattern matters more than the referral share. If the sources cited in your niche are directories and distributors, the facts buyers see about you may come from your listings there. Check those listings before writing new blog posts.
Buyer research points the same way. In a Gartner survey of 632 B2B buyers (August–September 2024), 61% preferred a rep-free buying experience overall. Yet for tasks that need context, such as judging whether a product fits their needs, buyers preferred seller input. And 69% reported inconsistencies between a supplier’s website and what its sellers said.
For manufacturers, that suggests a sequence. Let buyers check capability on their own, from pages that match what your engineers will say. Then make a fit review easy to request. Gartner also found 73% of buyers avoid suppliers who send irrelevant outreach. The handoff should answer the buyer’s request, not start a generic sales sequence.
Four observations, four different denominators
These figures measure different things, so they cannot be combined into a funnel. The table keeps each result next to its base. Sources: Semrush for the first three rows and Gartner for the last.
| Observation | What was counted | Result | What it cannot tell you |
|---|---|---|---|
| AI Overview presence | Search volume for 458 manufacturing keywords | 38% in January to 57% in July 2026 | Which sources were shown or clicked |
| AI referral share | US sessions to manufacturing sites | 0.48% from AI assistants and AI Mode | Visits made later, after reading an answer |
| Mention–citation overlap | Top 15 domains by mentions and by citations | 2 domains on both lists | Whether either result produced an inquiry |
| Buyer preference (Gartner) | Survey of 632 B2B buyers, August–September 2024 | 61% prefer rep-free buying; 69% saw website–seller inconsistency | Manufacturing buyers in particular; the sample covers B2B buyers generally |
Build a capability proof matrix
Worked example — a fictional contract CNC machining shop. Its pages promise “precision machining for demanding industries.” That phrase answers no engineer’s question and gives an AI system nothing specific to cite. The team lists the questions buyers ask before requesting a quote, then names the approved proof for each.
The matrix records sources and owners, not invented numbers. Where a figure is needed, it points to the approved document that holds it.
| Buyer question | Approved proof to publish | Owner | Do not claim |
|---|---|---|---|
| Can you machine our material? | Materials run in production in the last 12 months, taken from job records | Production manager | Materials quoted but never run |
| Can you hold our tolerances? | Capability statement by feature type, with the inspection method that verifies it | Quality manager | One headline tolerance for every part |
| Which inspection records will we get? | Sample inspection report with client details removed | Quality manager | Reports you do not routinely supply |
| Is your quality system certified? | Certificate scope, issuing body and expiry date, in plain words | Quality manager | Coverage beyond the certificate scope |
| What volumes and lead times fit? | Approved ranges, with the conditions that change them | Planning lead | Fixed lead times with no conditions |
| Who reviews our drawing? | The reviewing role, what to send and how the reply arrives | Applications engineer | Instant quotes you cannot honour |
| What drives the price? | Main cost drivers and quote validity, written for procurement | Sales operations | Price promises before drawing review |
The last row serves the economic buyer, often a procurement manager, who needs cost drivers and terms rather than machining detail. The other rows serve the engineer who must approve a new supplier.
The matrix has limits. It shows what the shop can do in general. It cannot confirm fit for a specific part; only a drawing review can. Say so on the page, and make the review request easy to find.
Define the inquiry before you measure it.
Agree what counts as a qualified inquiry before tracking anything. For the fictional shop, a drawing or specification is attached, the material is on the approved list, the quantity sits inside the published range, and the contact can approve a supplier or name who does. Log spam, tests and duplicates separately.
Use the same question set and period for mention and citation checks. Report qualified inquiries by landing page for that period separately; an inquiry may not reveal which AI question came before it. Do not add them up or treat one as proof of another. If owned citations rise but qualified inquiries do not, check whether the cited page leads to a clear fit review. If an answer cites a page for something you no longer offer, use the citation diagnostic.
GEO for Manufacturers: Common Questions
Should manufacturers check distributor and directory listings?
Yes, early. Third-party listings can describe your products to buyers and to AI systems. Correct your own pages first, then ask each listing owner to update outdated details through an authorized contact.
Does a low AI referral share mean we can ignore AI search?
No, but it does mean referrals are a weak sole measure. Visits after an AI answer can arrive as direct or branded search. Track mentions, owned citations and qualified inquiries separately, and do not credit AI with changes you cannot trace.
Can we publish tolerances and certifications?
Yes, when their owner approves them and the page states their scope. Tie each tolerance to a feature type and an inspection method. Show certificate scope and expiry, and remove anything out of date.
How does this differ from a buyer-question content plan?
A buyer-question plan decides which questions to answer and where. The capability matrix decides what proof each manufacturing answer needs and who signs it off. Most teams need both.
Will strong capability pages get us cited in AI answers?
Nobody can promise that. Clear, specific, approved facts give AI systems and buyers something useful to cite. Whether a given answer cites you depends on the platform, the question and the competing sources.
Find out which proof is missing.
If your brand appears in AI answers but inquiries do not reflect it, start with the evidence. An AI visibility audit reviews an agreed set of application questions: where your brand is mentioned, which sources are cited and which buyer questions lack approved proof. You receive a prioritized roadmap, and implementation is scoped separately. Bring your priority products, target markets and the competitors buyers compare you with.