What pages matter most if I can only fix five for AI visibility

If you can only fix five pages for AI visibility, fix the homepage, the one service or product URL that pays rent, About, a FAQ or proof page, and either location or contact. Those five are the URLs models already try to quote when someone asks who you are, what you sell, whether you are real, and how to reach you. A fifty-post calendar does not replace an extractable offer sentence and an entity page that agrees with it.

This spoke sits under the Answer Engine Optimization playbook. Lane work lives on /visibility. I have been SEO certified since 2021 and have shipped hundreds of production sites. The five-page bet is still a page-job decision, not a new file format.

The short answer

• Homepage: one-sentence who, what, for whom, and where — in HTML, above the fold.
• One service or product URL: the offer a buyer would type, with scope, constraint, and next step.
• About: legal name, people, location, founding facts, and the same offer language as the homepage.
• FAQ or proof: the objections and receipts a model will invent if you leave them blank.
• Location or contact: NAP (name, address, phone) plus how a human actually starts work — pick one as the fifth URL using the table below.

Why do five pages beat a fifty-page rewrite?

Answer engines compress the site into a handful of passages. Google’s own AI features documentation says a supporting link in AI Overviews or AI Mode must be indexed and snippet-eligible in ordinary Search. There is no extra AI markup track. Google’s generative-AI optimization guide (published May 2026) repeats the same floor and tells you to ignore llms.txt, AI-only schema, and pre-chunked “AI versions” of pages as Google Search tactics.

If only five URLs can be trusted this quarter, they should be the five a model needs to answer “who is this, what do they sell, can I trust them, how do I start.” Blog volume is a cluster problem. It is not the first five.

Bet	What you hope happens	What usually happens
Rewrite 50 blog posts	Coverage looks serious	Models still quote a vague homepage
Ship a new llms.txt	“We did AEO”	Google treats it as a text file; the five pages stay mush
Add AI-only schema	Secret eligibility	Schema that does not match visible text becomes a conflict
Fix the five-page set	Citeable who / what / proof / next step	Prompt answers stop inventing your offer
Expand a cluster first	Topical authority	You scale the wrong template — see content clusters for AI visibility after the five are honest

Operator rule: eligibility is a floor. Selection is a writing and entity problem. Fix the floor on five URLs before you fund a library.

What is the concrete five-page set?

Name the set out loud. If you cannot point at five live URLs, you do not have a plan. You have a slogan.

Slot	Default URL job	Primary question it must answer in the first 60–80 words	Citeable object
1. Homepage	Identity + offer + next step	Who are you, what do you sell, who is it for?	One-sentence offer + one table or three bullets
2. Service or product	The money page	What exactly do I buy, what is in / out of scope?	Scope table + CTA to contact, cart, or booking
3. About / entity	Disambiguation	Which legal entity, which people, which city?	Fact table that matches schema
4. FAQ or proof	Objections and receipts	What do buyers ask, what can you stand behind?	Q&A in HTML, or named proof — not adjectives
5. Location or contact	How to start	Where are you, how do I reach a human?	NAP + hours or form path, consistent with slot 3

That set is the default for a single-offer operator. Multi-brand companies still pick five per brand the model must not confuse, not five across the whole group.

Checklist before you rename the tickets:

• Each slot maps to one canonical URL (no “homepage-ish landing page” twins)
• The five URLs already return 200 and are indexable
• You can say the offer sentence the same way on slots 1, 2, and 3
• Slot 4 is either FAQ or proof, not a sixth blog
• Slot 5 is location or contact, chosen with the next table

If slot 2 is “all services,” you already failed the constraint. Pick the one that pays rent.

“Fix” is three jobs on the same URL, not a CMS redesign.

Job	Done means	Not done means
Eligibility	Indexed, snippet-eligible, answer in the fetched HTML	noindex, empty JS shell, nosnippet on the lead
Extractability	A 60-word lift names who, offer, and constraint	Atmosphere first; the answer lives on scroll four
Agreement	Same name, city, and offer as the other four URLs	Homepage says studio, About says agency, schema says the wrong city

If you only clear eligibility, you stay in the candidate pool and still get misquoted. If you only rewrite copy on a noindex URL, you wrote a private brochure.

How do I adapt the fifth slot — location or contact?

The first four slots stay. The fifth is a fork. Use the table. Do not “do both as half pages.”

Signal	Fifth URL is location	Fifth URL is contact
How buyers find you	Map pack, “near me,” city + category	Search, referral, LinkedIn, inbound form
Offer type	Local service, shop with a counter, studio visits	Remote service, SaaS, nationwide, digital delivery
Entity risk	Wrong city, missing NAP, hours that fight Google Business Profile	Form with no human path, no email, no response window
Example shape	AllCity HVAC-style contractor; a shop like Divine Toke with a real pickup story	Fractional work, remote builds, “book a call” studios
What the page must contain	Name, address, phone, hours, service area, map embed that matches the text	Named next step, who replies, what to send, timezone, no mystery inbox
When to swap later	You open a second city — then location stays fifth and contact becomes a sixth after the five are done	You add a flagship studio — then promote location; do not demote contact until NAP lives somewhere canonical

Decision procedure:

1. Write the query a stranger would type: city + category, or “hire / buy / book [offer].”
2. If the winning SERP is Maps plus a local pack, fifth = location.
3. If the winning SERP is a form, demo, or “talk to us,” fifth = contact.
4. If both fire (local brand with a strong inbound form), keep location as fifth and put a short NAP block on contact — one canonical address string.
5. Do not create /locations and /contact as competing entity pages in week one.

Artists and catalog brands (Fuller, a shop, a label) still need slot 5. Tour dates are not a substitute for a contact or a city the model can attach to the entity.

What do I actually change on the homepage?

Stop treating the homepage as a mood board. It is the most-fetched URL in most prompt logs I run, and it is the page most likely to be quoted as “what the company is.” I will not invent a citation-rate percentage for that. Treat it as the default identity document until your Search Console and prompt panel say otherwise.

Write it answer-first: 40–80 words that still work if a model lifts them with no hero video.

Element	Cut	Keep / add
First screen	Metaphor, reel, “we are passionate”	Who, offer, for whom, city or remote, one next step
Proof	“Trusted by industry leaders”	Named work you can defend — portfolio names without fake metrics
Nav	Twelve products	The money URL, About, FAQ or proof, location or contact
Media	Autoplay hero that delays text	Text in HTML; video optional after the sentence
Footer	Designer credit only	NAP or contact line that matches slot 5
Schema	Organization JSON-LD that names a different city	Organization (or LocalBusiness) that matches visible copy — schema.org Organization

Homepage procedure:

1. Draft the 60-word identity sentence on paper. Read it out loud. If you would not say it on a sales call, rewrite it.
2. Put that sentence in the first text node of the HTML, not inside a canvas or a delayed client island.
3. Add one table or three bullets: offer, constraint, next step.
4. Link slot 2, slot 3, and slot 5 in the first screen or the first scroll, not only in the footer.
5. Fetch as Googlebot. If the sentence is missing, the page is not done.

Homepage checklist:

• Identity sentence visible with JS off, or in the first server HTML
• Same legal name as About and schema
• One money-page link with the same offer words
• No nosnippet or max-snippet:0 on the lead
• Screenshot of URL Inspection: indexed, snippet-eligible

60-word quote test (run before design QA):

Test	Pass	Fail
Hide everything after the first 60 words	You know who, offer, for whom	You know a mood
Paste those 60 words into a blank doc	They still name the city or say remote	They need the logo to make sense
Compare to the money-page lead	Same offer words	Synonym soup

A beautiful homepage that cannot be quoted is a brochure. Bravery is not a citation strategy.

Which service or product URL is the money page?

Pick one. The money page is the URL a qualified buyer should land on after the model names you. It is not your sitemap. It is not “Services” as a teaser grid.

Candidate	Keep as slot 2 when	Demote when
Flagship service	It is the offer you would take a call about this month	It is a legacy SKU you no longer sell
Flagship product	It is the SKU that funds the business	It is a merch afterthought
Pricing	Price is the question and you will publish a real range	Price is a bait table that fights the sales team
Category hub	You are a shop and the hub is how people buy	The hub is a thin grid with no scope language
Case study	Never as slot 2	Proof belongs in slot 4

If you run three real offers, still pick one for the five-page sprint. The other two wait. Conflicting service pages are how models invent a package you do not sell.

Money-page procedure:

1. List offers that produced pipeline or revenue in the last two quarters. No vibes.
2. Choose the one you would staff next week.
3. Write a scope table: included, not included, who it is for, who it is not for, how to start.
4. Put the next step on this URL (audit, cart, booking). Do not bounce the only CTA to a newsletter.
5. Repeat the offer sentence from the homepage. Synonyms here are not clever. They are entity drift.

Block	Must contain	Failure if missing
Lead	Offer + constraint in 60–80 words	Model quotes the homepage metaphor instead
Scope table	In / out / buyer / non-buyer	Hallucinated deliverables
Proof chip	One named receipt or constraint, not a percentage you cannot source	Fake social proof
CTA	The same verb as slot 5	Dead-end “learn more”
Internal links	Home, About, FAQ or proof	Orphan money page

I will cite receipts I can defend: 500+ automations built, 20,000+ hours on agentic systems, 35,000+ hours saved for clients, hundreds of sites shipped. I will not staple a fake conversion rate to a client name. If the money page needs a number, use a number you would repeat on a call.

What belongs on About so the entity stops hallucinating?

About is not a brand poem. It is the disambiguation page. Models mash similar names, old addresses, and former taglines unless you publish a fact table and keep it consistent with Organization or LocalBusiness markup.

Google’s people-first content guidance still applies: show who created the content and why it is trustworthy. That is an About job, not a blog-author byline job alone.

Fact	Publish on About	Also match on
Legal / brand name	Exact string	Homepage, schema name, slot 5
What you sell	Same offer sentence as slot 2	Homepage lead
People	Who is accountable, with roles	Author bylines if they exist
Place	City / region, or explicit remote	Location page or contact
History	Founding year only if true	Press kits, LinkedIn, GBP
Contact path	Link to slot 5	Footer

About checklist:

• Fact table in HTML (not a PDF)
• No second founding year hiding in the footer
• Same city string as Google Business Profile if you have one
• Same offer words as the money page — no “we do everything”
• Organization or LocalBusiness JSON-LD that copies the visible table, not a designer’s placeholder
• A human the model can name without inventing a co-founder

Procedure when the model already has you wrong:

1. Collect five prompt answers: “Who is [brand]?” “Where is [brand] based?” “What does [brand] sell?”
2. Highlight every invented fact.
3. Put the correction in the About table and the homepage sentence. One page is not enough if the other still teaches the old line.
4. Request indexing on both URLs after the HTML matches.
5. Re-run the same five prompts in two weeks. Do not declare victory from one ChatGPT session.

If About is a slider of values adjectives, you trained the model to improvise. That is the failure.

How should FAQ or proof pages earn the fourth slot?

Slot 4 exists because models fill silence. Pricing boundaries, turnaround, who it is for, and “are you the same company as X” get invented if you will not say them.

Pick FAQ when the questions are stable and the answers are policy. Pick proof when the risk is “are they real,” and you have named work you can stand behind. Do not burn the slot on a generic “FAQ + testimonials” wall of stars.

Slot 4 type	Use when	Citeable unit	Do not
FAQ	Buyers repeat the same five to eight questions	H2/H3 questions with answers in HTML	Accordion content missing from the fetched document
Proof / work	Trust is the blocker	Named project, constraint, what shipped	Invented percentages on portfolio names
Brand FAQ hybrid	Entity collisions (same name, two companies)	“We are / we are not” table	A joke page that never states the legal name
Pricing FAQ	You will publish a real range or “starts at” you honor	Range + what changes the number	A range the sales team will contradict

Google does not require FAQPage schema to appear in AI features. schema.org FAQPage is still fine when it matches visible Q&A. It is not a cheat code, and FAQ rich results in classic Search are limited. Write for the HTML.

FAQ / proof checklist:

• Every answer is in the initial HTML
• Each Q&A could be lifted in four sentences or fewer
• Proof names a real project without attaching a fake metric
• Policy answers match the money page (refunds, scope, timeline)
• “Are you hiring / is this a job?” does not crowd out buyer questions
• Internal links back to slots 1, 2, and 5

If you cannot name five honest questions, you are not ready for slot 4. Write the sales-call transcript first, then the page.

What does the location or contact page need to be citeable?

Slot 5 is where invented phone numbers, dead emails, and the wrong city go to die. Treat it as a structured fact page, not a contact-form decoration.

Field	Location page	Contact page
Name	Matches About and GBP	Matches About
Address	Full NAP, one canonical string	NAP if you have a place; otherwise city + remote
Phone	Click-to-call, same as GBP	Optional; if present, same as everywhere
Hours	Visible text, not only the embed	Response window (“replies in one business day”)
Service area	Cities / radius you will actually serve	Timezone (America/New_York for this studio)
Next step	Directions, call, or book	Form fields you will read; intent if you use one
Schema	LocalBusiness matching the visible NAP	Organization + ContactPoint matching the visible path

Location / contact checklist:

• One NAP string copied into About, footer, and schema
• Hours or response window in text
• Form works without a third-party modal that search cannot fetch
• No competing /get-in-touch clone with a different phone
• Map embed (if any) does not contradict the address line
• Intent query on the form if you run audits (intent=visibility-audit is a contact pattern, not a homepage pattern)

Procedure:

1. Dump every phone, address, and inbox from the site, GBP, LinkedIn, and the footer.
2. Pick one of each. Redirect or delete the rest.
3. Publish slot 5 with the winning set.
4. Re-fetch. Confirm the facts are in the HTML.
5. Only then touch schema.

A contact page that is a mailto: and a hope is not a fifth page. It is a leak.

What should I skip if I only have a week?

A week is enough to make five URLs honest. It is not enough to rebuild information architecture, launch a cluster, or “do AEO” as a campaign.

Skip this week	Why	Do instead
New blog calendar	Volume without extractable identity	Five URLs, answer-first leads
llms.txt / /ai markdown mirrors	Google’s May 2026 guide says not to treat these as a Search pathway	Spend the hours on HTML the crawler already fetches
AI-only schema products	Not required for Overviews / AI Mode	Organization / LocalBusiness that matches the table
Rewriting every service URL	You will ship five half pages	One money page
Vendor AEO scores	Not your P&L	Prompt panel + Search Console
Hero video replacement as the project	Media is not the citation unit	Text layer first
Merging the entire blog	Cluster work comes after identity	Bookmark the cluster spoke; do not start it

Week checklist:

• Five URLs named in a doc with owners
• Identity sentence locked
• Scope table on the money page
• About fact table
• Slot 4 Q&A or proof in HTML
• Slot 5 NAP or contact path
• URL Inspection on all five
• Prompt panel of 15 queries saved (same wording, same tools)

If someone adds ticket six, it waits. That is the whole point of five.

What fails first when teams try this?

The failure is almost never “we picked the wrong CMS.” The failure is picking the wrong five, or spreading one sentence across forty URLs and calling it a rewrite.

Failure	What it looks like	Cost	What you do instead
Five blog posts as the five	“We optimized our best rankers”	Models still invent the offer	Swap in home, money, About, proof, location/contact
Homepage essay	2,000 words, no offer sentence	The lift is paragraph twelve	60-word lead + table
About vs schema fight	JSON-LD city ≠ footer city	Hallucinated headquarters	One fact table, then markup
FAQ in a client-only accordion	Googlebot gets empty <div>s	Slot 4 does not exist	Server-render the answers
Two money pages	“Websites” and “automation” both claim to be the offer	Model averages them into a package you do not sell	One slot 2; the other waits
Contact twins	/contact and /book with different emails	Lost leads, wrong citation	One canonical next step
Fake receipts	Invented % on a client	You cannot defend the quote	Named work, no number, or a number you own
Skipping index checks	Beautiful HTML, noindex leftover	Zero eligibility	AI features eligibility first

Concrete failure mode: a studio spends the week rewriting five thought-leadership posts. The homepage still says “digital experiences” and the service page still says “we do it all.” ChatGPT and Perplexity keep describing a generic agency in the wrong city. Search Console never gets a chance to show generative-AI impressions on the URLs that matter because those URLs never became quoteable.

I have watched that pattern on more than one of the hundreds of sites in the book of work. I will not hang a fake recovery percentage on it. The fix is always the same: name five URLs, make them agree, then measure.

How do I measure whether the five-page bet is working?

Do not buy a score. Freeze a prompt panel and watch Search Console.

Google launched a Search generative AI performance report in June 2026 for a subset of properties. It is impression-led. If you do not have it yet, do not invent a workaround metric. Use the Web performance report, URL Inspection, and a manual citation log.

KPI	Source	Cadence	Pass	Fail
Eligibility	URL Inspection on the five	Day 1 and after each ship	Indexed, snippet-eligible, live HTML has the lead	noindex, nosnippet, empty fetch
Prompt accuracy	15 frozen queries across Google (AI Overview / AI Mode when present), ChatGPT, Perplexity	Weekly for 8 weeks	Offer, city, next step match the five pages	Invented SKU, wrong city, dead phone
Citation / mention	Same 15, screenshot	Weekly	Your URL or correct name appears	Competitor or a dead domain
Generative-AI impressions	Search Console report if present	Weekly	Impressions on the five URLs, not only old blog posts	Report exists and the five are absent for a full month — then check eligibility again
Brand queries	Search Console	Weekly	Not in free-fall	You “focused” by gutting identity pages
Next-step conversions	Analytics / CRM on slot 2 and slot 5	Weekly	Form, call, or cart events hold	Sessions on the blog were the only slide

Prompt panel (minimum 15, same wording every week):

1. Who is [brand]?
2. What does [brand] do?
3. Where is [brand] located?
4. How do I contact [brand]?
5. What does [brand] charge / how does pricing work?
6. Is [brand] the same as [confusable name]?
7. Who is [person]?
8. Does [brand] offer [money-page service]?
9. [City] [category]
10. Best [category] for [buyer]
11. [Brand] vs [obvious competitor] — only if you will publish an honest comparison later
12. [Brand] reviews / is [brand] legit?
13. [Brand] hours / service area (local) or [brand] timezone / booking (remote)
14. How to start a project with [brand]
15. [Exact offer sentence as a query]

Do not mash those tools into one “AI visibility %.” They are different products.

Surface	What a cite looks like	Hedge you keep
AI Overviews / AI Mode	Supporting link in the unit	Impression is not a session; Google does not guarantee inclusion
ChatGPT (when sources show)	Named source or URL chip	The UI moves; screenshot the date
Perplexity	Numbered source	Same query, different day, different set — log it

Semrush-style SERP views help you see classic blue links. They do not replace the prompt log. Use them as a packing list, not a grade.

Measurement checklist:

• Panel saved in a sheet with date, tool, cited URL, invented facts
• Same device class noted (desktop vs mobile Overviews can differ)
• Five URLs listed as the only “priority pages” in the weekly review
• No new KPI invented mid-month

If accuracy on questions 1–4 is still wrong after the five URLs are indexed with matching HTML, you have a conflict off-site (GBP, Wikipedia, a marketplace). That is a next sprint. It is not a reason to open the blog calendar.

What is the one-week ship order?

Do the pages in this order. Homepage last among the rewrites only if identity is already honest; otherwise homepage first. Default order below assumes the homepage is currently a mood board.

Day	URL	Done looks like
1	Triage + slot 5 facts	One NAP / inbox list; fifth slot chosen
2	About	Fact table live; schema matches
3	Money page	Scope table + CTA live
4	FAQ or proof	HTML Q&A or named proof live
5	Homepage	60-word lead + links to 2, 3, 5
6	Slot 5 page	Location or contact shipped; twins redirected
7	Measure	Inspection on five; prompt panel v0 screenshots

Numbered procedure:

1. Export the current homepage, money page, About, FAQ/proof, and contact/location to a folder. You need a before.
2. Lock the identity sentence and the offer sentence. They can be the same sentence.
3. Ship About and slot 5 facts first if the model already has the wrong city. Entity errors poison the rest.
4. Ship the money page with a scope table. Link it from Home and About in the same deploy.
5. Ship slot 4. Kill accordion-only answers.
6. Rewrite the homepage lead. Remove the metaphor. Keep the craft in the layout if you want — after the sentence.
7. Request indexing. Run the 15 prompts. File invented facts as tickets against the five URLs, not against new posts.

Owners:

Role	Owns	Does not own this week
Operator / founder	Identity and offer sentences	A new cluster outline
Designer	Text visible without breaking layout	A new art direction
Developer	Server HTML, schema match, redirects	An AI crawler product
SEO	Inspection, prompt log, Search Console	A 40-post brief

If day 7 still has two offer sentences, you did not ship. You drafted.

When is this not worth doing yet?

Skip the five-page sprint when the site cannot be a source. Eligibility and a real offer come first.

Blocker	Why the five pages will not save you	What to do first
Sitewide noindex or login wall	Not snippet-eligible	Remove the accident; read AI features
Domain not in Search Console	You cannot measure	Verify the property
Legal name in flux (rebrand this month)	You will teach two entities	Freeze the name, then write
No offer you will take	Slot 2 is fiction	Pause marketing pages
Duplicate sites (old Webflow + new Next)	Models cite the zombie	301 the zombie, then fix five on the survivor
Someone else owns the brand search	Marketplace or franchise page outranks you with different NAP	Align off-site facts or pick a disambiguator
You cannot staff a reply to slot 5	Contact citations create angry humans	Set a response window you will keep

Checklist — proceed only if:

• You can name the offer you will take next week
• The five URLs are on the canonical domain
• You have Search Console (or will verify this week)
• You can edit HTML, not only a tag manager banner
• One person owns the identity sentence

If those boxes are empty, a visibility audit is the right next step, not a content calendar. The lane page is /visibility. The booking path is a visibility audit.

The five-page set is a constraint. Treat it like one.

FAQ

What pages matter most if I can only fix five for AI visibility?

Homepage, one service or product URL that pays rent, About, a FAQ or proof page, and location or contact. Those five cover identity, offer, entity, objections, and how to start — the questions models already try to answer about you. Everything else waits until those URLs agree in HTML.

How do I measure whether fixing five pages for AI visibility is working?

Run a frozen 15-prompt panel weekly and score offer, city, and next-step accuracy, plus whether your URL is cited. In Search Console, confirm the five are indexed and snippet-eligible, and use the generative-AI impressions report if your property has it. Do not treat a vendor AEO grade as the scoreboard.

What usually fails first when teams try this?

They pick five blog posts, or they rewrite the homepage into a long essay with no offer sentence. The next failure is entity drift: About, schema, and the footer disagree on city or name. Fix agreement on the five URLs before you add a sixth.

How long does this take to show results?

Eligibility can be checked in days with URL Inspection once the HTML is live. Accurate answers on a frozen prompt panel often take weeks, and citations are not guaranteed — Google’s docs say indexing and serving are never a promise. Re-measure at two weeks and eight weeks on the same 15 queries; do not invent a day-count SLA.

What should I skip if I only have a week?

Skip a new blog calendar, llms.txt theater, AI-only schema, and rewriting every service page. Spend the week on the five URLs: identity sentence, scope table, About facts, HTML FAQ or proof, and one location or contact page. Index those five. Then stop.

When is this not worth doing yet?

If the site is not snippet-eligible, the domain is not in Search Console, the legal name is mid-rebrand, or you do not have an offer you will actually take. Fix eligibility and the offer first. The five-page rewrite cannot cite a product that does not exist.

CTA

If you can only touch five URLs, make them the ones a model must get right — then book the audit if the prompt log still lies.

Lane: /visibility · Book a visibility audit.