The pages I open first when a brand wants AI answers to stop skipping them

Last Thursday I pulled up a fashion label’s site the way an answer engine would. I skipped the human browse path. No slow scroll through the lookbook. I went straight to About, then Contact, then the one page that claimed results, then the product collection that was supposed to be their hero.

About was three sentences of mood. Contact was a form with no street address and a Gmail address. The results page said “trusted by thousands” with no names, no dates, and no numbers you could check. The hero collection had beautiful photography and almost no plain facts about materials, origin, or who makes the clothes.

A person might still fall in love with the product. An answer engine has much less to hold onto. When someone types a buying or hiring question into ChatGPT, Perplexity, or a Google AI Overview, the systems that answer do not invent a favourite. They assemble sources they can defend. Thin pages get skipped. Specific pages get quoted.

That is the practical half of generative engine optimisation that most product brands still treat as optional: source readiness. The full method lives on that pillar. This piece is about the pages I check first, and what has to be true on them before you have a fair shot at being named.

What “getting cited” means for a product business

Getting cited is a commercial moment: an answer engine uses your page as a source when it answers a real question — who makes a durable linen set in Australia, which agency runs coordinated ecommerce marketing, what a marketing diagnostic includes, which brand stocked a certain wholesaler.

For the owners I work with, the offline business is already real. Physical retail. Wholesale accounts. A product people reorder. The online side is often thinner than the product deserves: a Shopify theme, occasional posts, ads that switch on for a launch. When AI answers start shaping discovery, that thinness becomes expensive. Your competitor with plainer copy and clearer proof can get named while your better product stays invisible in the answer box.

I wrote earlier about how your next customer might not be a person in the first pass of research. The follow-on question is simpler. If a model did open your site, would it find anything solid enough to cite?

The ten-minute pass I run on a site

I do not start with tools. I start with four URLs and a notebook.

About. Does it name a real person, a real place of business, and a real operating history? “Based in Australia, passionate about quality” is atmosphere. “Founded in 2014 on the NSW coast, wholesale to 40 stockists, retail online since 2019” is something a model can keep straight across pages.

Proof. Is there a case studies style page, or client results with dates, scope, and numbers? Vague testimonials without context are weak. A short write-up with a before state, what changed, and a result you would defend in a meeting is strong. Public Clutch reviews help for service businesses for the same reason: a third party recorded the claim.

Commercial clarity. Can a stranger tell what you sell, who it is for, and how to buy or book within one screen? Product brands hide this behind lifestyle copy. Agencies hide it behind service poetry. Both patterns cost citations.

Consistency. Do the same facts appear the same way on About, footer, Google Business Profile, and any press page? Answer engines punish contradiction quietly. They simply prefer the source that agrees with itself.

If those four fail, I rarely jump to schema markup or keyword lists. Markup on a hollow page is decoration. The order is substance, then structure, then distribution.

Claims a model can verify

Models prefer statements that look checkable. Skip the lab-report standard on every line. Write the version of the truth that would survive a sceptical reader.

Weak: “We deliver exceptional results for growing brands.”

Stronger: “Since 2019 we have run paid ads, email, SEO content, and site maintenance for a fashion and homewares brand with wholesale and retail channels. Online retail sits around $50K a month on the current setup.”

The second version can be wrong or right. The first version cannot even be tested. Answer engines are built to lean toward the testable line when they have a choice.

I apply the same standard on McKee Creative’s own pages. The Brand System Diagnostic inclusions page exists so a buyer — or a model summarising for a buyer — can quote fee, scope, timeline, and boundaries without guessing. The Marketing Diagnostic booking page is where that work starts. Separate pages for “what it is” and “book it” is deliberate. Citation-friendly pages answer. Conversion pages convert. Mixing both into one foggy URL usually does neither job well.

Product brands can copy the pattern without copying the offer. One page that states materials, make, shipping regions, and wholesale terms in plain language will outperform three mood boards when a model has to pick a source.

Where product brands usually under-build

Fashion, homewares, wellness, specialty food and bev — the categories I know best — often invest in photography and under-invest in the boring layer.

Collection pages that never name fibre content, care, or country of make.

About pages that never name the founder or the workshop.

Stockist pages that list logos without cities.

Launch emails that make a claim the product page never repeats.

Press mentions that live only on Instagram stories.

None of that matters much if every sale still comes from a wholesale appointment or a local walk-in. It matters a great deal when a buyer asks an AI which Australian brand fits a brief, or which maker supplies a certain category. The model looks for pages it can quote without embarrassment. Mood copy rarely survives a citation. Plain facts do.

Wholesale-strong businesses have an advantage here if they use it. Stockist lists, trade-show history, and reorder patterns are proof. Put them on the site in calm language. Do not leave the strongest evidence offline.

How this ties to hire and buy questions

Citation work covers product discovery and hire questions. Owners ask answer engines who to hire as often as they ask which product fits a brief. I covered that angle in when a customer asks an AI who to hire. The same source-readiness rules apply to agencies and to the brands those agencies serve.

If you sell a physical product, you want the model to find your category facts and your proof.

If you are choosing a marketing partner, you want the model to find coordinated delivery, documented outcomes, and a clear next step — not a homepage full of slogans.

When I rebuilt how McKee Creative publishes, I treated public proof as part of the marketing function, not a side project. Case studies, plain offer pages, and specific method pages are how a solo operator with an in-house AI marketing system stays legible to both humans and machines. The AI marketing agency conversation is downstream of that. Labels are cheap. Readable operations take longer to build.

A working order for the next fortnight

If you want a concrete sequence rather than a philosophy, use this.

Week one: rewrite About with name, place, history, and what you sell. Align the footer and Google Business Profile to the same facts. Add one proof page with a single documented outcome you will stand behind — or publish a short project write-up with dates and scope if you sell services.

Week two: pick your top three money pages (best collection, best service, or the offer you most want inbound for). Add the checkable claims those pages are missing: who it is for, what is included, price range or path to price, shipping or delivery regions, and one paragraph a stranger could quote without calling you first.

Then stop and re-read the site cold. If you cannot find a sentence worth citing, neither can an answer engine.

Broader generative engine optimisation work — topical clusters, support articles, technical crawl paths, ongoing content — sits on top of that base. Clean the four pages first, then build the wider content program. The wider program fails quietly when the base is empty.

When to get a full read instead of patching pages

Some sites need more than a fortnight of page repairs. Channels fight each other. Ads point at one offer while email pushes another. The blog ranks for questions you no longer want to answer. The owner is still the unpaid editor of every channel.

That is the moment I use the Marketing Diagnostic rather than another isolated fix. Fixed fee. Four pillars: paid ads, email, SEO and content, and how those pieces coordinate. Written priorities you can hand to a future hire or keep in-house. If the map says the next step is done-for-you delivery, the Economy Marketing Package is the ongoing retainer built for that. If the map says fix proof and entity clarity first, you leave with that list and no obligation to stay.

For owners who want to see how the work looks in public before they book anything, start on the case studies page. Read the numbers. Then decide whether you need a diagnostic or a fortnight of page work you can do yourself.

If Thursday’s traffic still arrives from wholesale contacts and walk-ins, you can delay this. If buyers already ask tools who makes what you make, the four pages are overdue.