What belongs in an AI visibility monthly retainer vs a one-time audit

A one-time AI visibility audit is a baseline plus a prioritized fix list. A monthly retainer is the work that does not finish at the readout: prompt-panel tracking, entity hygiene, page-job rewrites, and citation recovery. If you buy a retainer to get a second PDF of the same findings, you bought a report subscription. If you buy an audit and expect ChatGPT or Google AI Overviews to start citing you next Tuesday with nobody shipping HTML, you bought a diagnosis and called it treatment.

This spoke sits under the Answer Engine Optimization playbook. Definitions live in what AI visibility actually is. How to score the work lives in measuring AI search visibility. Lane work is on /visibility. I have been SEO certified since 2021. I will not invent a monthly fee or a client citation-recovery percentage in this post. The object you buy is the decision. The price is a separate conversation.

The short answer

• Audit: freeze the current state, name eligibility blockers, rank the fixes, hand you a 30/60/90. Then it stops.
• Retainer: run a frozen prompt panel, keep entity facts aligned, rewrite page jobs, and recover citations after losses. That is weekly labor, not a recap.
• Buy the audit first if you cannot answer “are we even eligible, and which five URLs matter.”
• Buy the retainer when the ranked list exists and the failure mode is “nobody ships.”
• Do not judge either object on restored 2022 informational CTR. Judge citations, brand search, and conversions on protectable URLs.

You are buying	Done looks like	Wrong cousin
One-time audit	Baseline + prioritized fixes, dated, owned	A content calendar disguised as diagnosis
Monthly retainer	Prompt-panel tracking, entity hygiene, page jobs, citation recovery	A monthly reprint of the audit PDF
Neither yet	Eligibility and Search Console exist; offer is real	Paying for screenshots of someone else’s niche

The audit finds the work. The retainer is the work. Mixing those two sentences is how retainers turn into theater.

What belongs in a one-time AI visibility audit?

A baseline you can defend, plus a ranked list of fixes you can staff. Not a 40-post outline. Not a vendor “AEO grade.” Not a promise that Google will cite you.

The audit answers four questions in writing: Are we eligible to be quoted. What do answer engines currently say about us. Which URLs can carry the answer. What do we do first, second, and not at all this quarter.

Audit artifact	What it contains	What it is not
Eligibility pass	Indexing, snippet rules, noindex / nosnippet accidents, robots, canonicals	A crawl of the whole internet
Prompt-panel v0	25–40 frozen buyer prompts, dated screenshots, mention vs citation vs absent	One ChatGPT chat from a sales call
Entity map	Legal name, NAP, offer sentence, About vs schema vs footer vs GBP if you have one	A brand workshop
Page-job list	Protect vs reinvent vs demote, with owners	A blog calendar
Prioritized fixes	Ranked tickets: page, fact, schema, corroboration, technical	“Do AEO” as a vibe
30/60/90	Sequence with a kill criterion	A retainer proposal stapled to the back

Google’s AI features documentation still says a supporting link in AI Overviews and AI Mode must be indexed and snippet-eligible. There is no special Overview markup. An audit that skips URL Inspection is a mood board.

Checklist — the audit is complete only if:

• Frozen prompt set written down (same wording next month)
• Each priority prompt labeled mentioned / cited / absent / wrong-fact
• Eligibility blockers named with the URL that fails
• Five or fewer first-wave URLs, not fifty
• Each fix has an owner and a “done looks like”
• You can read the ranked list out loud in five minutes

If the deliverable cannot be staffed without a translator, it is not a baseline. It is a brochure.

A ranked fix list should look like tickets, not themes. Example shape — not a client result, not a promise these pages will get cited:

Rank	Ticket	Done looks like	Owner
1	Homepage nosnippet accident	Answer block snippet-eligible; Inspection clean	Developer
2	About vs schema city mismatch	One city in HTML and JSON-LD	Developer + operator
3	Money page has no scope table	Citeable inclusion/exclusion table live	Content + operator
4	Comparison URL is a listicle	Criteria table + handoff to pricing	Content
5	Identity prompts invent a package	Offer sentence matches pricing HTML	Operator

Refuse these as “audit deliverables”:

Vendor line	Why it is not an audit
Six months of “implementation included”	That is a retainer wearing the audit’s name
A 40-post editorial calendar	That is content ops, and it usually skips the five URLs that matter
An “AEO score” with no prompt list	You cannot re-run it
A citation guarantee	Nobody honest can sign it
A monthly fee quoted in the audit deck	Price the object after you know the object

What belongs in an AI visibility monthly retainer?

Four jobs that recur. Prompt-panel tracking. Entity hygiene. Page jobs. Citation recovery. If a line item does not map to one of those four, it is classic SEO, design, or reporting — label it honestly or cut it.

A retainer exists because answer engines rewrite themselves. Same prompt, different day, different sources. A one-shot screenshot is an anecdote. A panel is a ritual.

Retainer workstream	Monthly output	Kill it when
Prompt-panel tracking	Same 25–40 prompts, scored mention / citation / accuracy, with SERP or answer screenshots	The prompt set changes every week so nothing trends
Entity hygiene	Diff of name, city, offer, pricing facts across About, schema, footer, profiles	You “rebrand copy” without updating schema
Page jobs	Shipped HTML on ranked URLs: answer-first lead, one citeable table, handoff to a protectable URL	You open twenty outlines and ship zero
Citation recovery	Tickets for lost citations, wrong facts, and competitor takeovers on money prompts	You only log the loss and never change a page

What the retainer may also include, if you name it as SEO rather than magic:

Adjacent work	Keep if	Do not relabel as AEO
Technical SEO floor	Indexing, speed, canonicals still break citations	“AI crawl optimization” with no Inspection
Classic rank tracking	Money queries still have blue links	Using rank as the only scoreboard
Content ops	You are shipping the ranked pages	A volume calendar to “outrank the model”
Digital PR / corroboration	A fact needs a second source	Buying links and calling them citations

Operator rule: if the monthly email is screenshots with no shipped URL, you are renting a report. Reports do not get you cited.

A month of retainer labor, if it is real, looks like a calendar — not a vibe.

Week	Tracking	Hygiene	Page jobs	Recovery
1	Full panel run; red cells filed	Diff Home / About / money / schema	Ship or explicitly defer URL A	Confirm last month’s losses on a second capture
2	Spot-check money prompts only	Fix the one fact that drifted	URL A indexed; start URL B	Smallest HTML change for a lost citation
3	Full panel	Recheck external profiles you control	Ship URL B	Re-run the recovered prompt
4	Monthly rollup: mention vs citation vs shipped changelog	Freeze sentences if a campaign wants to “improve” them	Changelog to the operator	Kill or keep each open recovery ticket

RACI you can paste into the SOW:

Job	Agency / freelancer	In-house operator	Developer
Prompt panel	Runs and scores if hired for it	Owns the prompt list; vetoes new prompts	None, unless a tool must be instrumented
Entity hygiene	Diffs and tickets	Freezes identity and offer sentences	Ships schema and NAP
Page jobs	Drafts the citeable object	Approves the offer and the CTA	Ships HTML, not a tag-manager banner
Citation recovery	Logs the loss; proposes the smallest change	Decides protect vs reinvent	Inspection and deploys
Classic SEO floor	Only if labeled SEO	Still your Search Console	Still your canonicals

If the SOW has a RACI with three empty columns, you hired a narrator.

How do you decide audit vs retainer without a price slide?

Start from what you cannot currently name. If you cannot name eligibility, current answers, or the first five URLs, you need an audit. If you can name all three and the pages still have the old job, you need a retainer. Price comes after the object. I will not invent a monthly number here.

Decision table:

Question	If no	If yes
Can you show a dated prompt log for 25+ buyer prompts?	Audit — you have no baseline	Maybe retainer — the log has to keep moving
Can you name eligibility blockers with URLs?	Audit	Retainer can keep Inspection in the weekly loop
Do you have a ranked fix list with owners?	Audit	Retainer ships the list
Has anyone shipped HTML on those URLs in 30 days?	Audit if the list is fiction; retainer if the list is real and idle	Retainer — the bottleneck is labor, not knowledge
Are answer engines inventing your city, price, or offer?	Audit to map the conflict	Retainer — entity hygiene is weekly
Did you lose a citation on a money prompt this month?	Audit if you never measured	Retainer — citation recovery is the job
Is the site not snippet-eligible or not in Search Console?	Neither. Fix eligibility first. Then audit.	Proceed

Procedure:

1. Write the four questions you actually care about buyers asking. If you cannot, you are not ready for a retainer.
2. Run those four prompts today, manually, on the surfaces you sell against (Google AI Overviews / AI Mode, ChatGPT, Perplexity — pick the ones your buyers use).
3. If you are surprised by the answers, buy the audit. Surprise is the baseline you do not have.
4. If you are not surprised, and the same five URLs have been “on the list” for a quarter, buy the retainer or staff it in-house.
5. If the site is noindex, login-walled, or missing from Search Console, stop. Eligibility before either object.

Failure mode: a founder buys a retainer because a competitor “does AEO,” then receives a dashboard login and no page diffs. That is not a strategy miss. That is buying the wrong SKU.

When is a one-time audit enough?

When the constraint is knowledge, not cadence. An audit is enough if a competent in-house person can take the ranked list and ship it, and you only need an outside hammer to settle the argument.

Situation	Audit is enough	You still need a retainer
First time measuring AI visibility	Yes — you need a baseline	After the list exists and week four still has zero HTML
Executives want an external ranked list	Yes — politics, not a year of agency	If nobody on staff will own the prompt panel
One eligibility mess (nosnippet, canonical loop)	Often yes, plus a developer week	Only if the mess recurs across templates
Entity facts already agree	Audit can be thin	Retainer still for panel + page jobs
You publish quarterly and can staff two URLs a month	Audit + internal ops	Retainer if that staffing is imaginary
Competitors own every category answer	Audit names the gap	Retainer is how you contest it month after month

Checklist — stop at the audit when all of these are true:

• Search Console is verified
• Prompt-panel v0 exists and will be re-run by a named owner
• Ranked fix list is five to fifteen tickets, not a novel
• Someone on payroll can edit HTML, schema, and internal links
• Leadership accepted protect vs reinvent vs demote
• You have a date for the first two shipped URLs

An audit that “includes three months of implementation” is a retainer with a costume. Split the SOW. Pay for diagnosis. Then pay for shipping, or staff it. Mixing them hides which one failed.

When does a retainer start earning its keep?

After the ranked list exists, and the failure is time. Answer engines drift. Entity facts rot. Citations disappear without a ticket. A retainer earns its keep when those four workstreams would otherwise die between other jobs.

It does not earn its keep as “someone to blame when ChatGPT is rude about us.” Models will be wrong. The retainer is whether you notice in a week and change a page, or notice in a quarter and rewrite the homepage in a panic.

Signal	Retainer is justified	Still an audit / DIY problem
Prompt answers change week to week on money queries	Tracking is a job	You have never frozen the prompts
Schema, GBP, and About disagree after every campaign	Entity hygiene is a job	One cleanup would actually stick
Informational URLs still written as click farms	Page jobs are a job	You have not decided protect vs reinvent
You were cited, then replaced	Citation recovery is a job	You never logged citations in the first place
In-house SEO is at capacity on classic rank + tech	Extra hands on the four workstreams	You have no SEO owner at all — hire or audit first
Leadership wants a monthly scoreboard that is not CTR	Ritual plus tickets	They want a fake ROAS number

I will cite receipts I can defend: SEO certified since 2021, 500+ automations built, 20,000+ hours on agentic systems, 35,000+ hours saved for clients, hundreds of production sites. I will not invent “retainers recover X% of citations in 90 days.” If a vendor says that, ask for their frozen prompt set and their definition of a citation. Then wait.

What does prompt-panel tracking include month to month?

The same prompts, the same scoring columns, a dated artifact, and tickets. Not a new tool login. Not a share-of-voice percentage invented from one screenshot. The measurement spoke is the full scoreboard; this section is what you are paying a retainer to do with the panel.

Procedure every week (or every other week if the set is large — never “whenever we remember”):

1. Open the frozen list. Do not add five trendy prompts and drop five money prompts.
2. Run each prompt on the surfaces you committed to. Same device class notes. Dated capture.
3. Score four columns: mentioned, cited (URL credited), accuracy of the fact, competitor set present.
4. File a ticket for every regression: lost citation, new wrong price, new wrong city, new competitor occupying the answer.
5. Ship or explicitly defer the ticket. A sparkline with no owner is theater.

Panel column	Pass	Fail
Prompt text	Identical to last run	Rewritten so the trend is fake
Surface	Named (AI Overviews, AI Mode, ChatGPT, Perplexity)	“AI” as one bucket
Mention	Brand or product named	You squint at a synonym
Citation	Your URL or a clearly credited source	A mention without a source treated as a link
Accuracy	Offer, geo, pricing match the live site	“Close enough”
Ticket	URL + owner + due date	“Monitor”

Google’s Search generative AI performance report (June 2026) is impression-led for AI Overviews and AI Mode, rolling out to a subset of properties, and it does not replace the panel. It does not show ChatGPT. If your property does not have the report yet, do not invent a workaround metric. Keep the manual log.

Checklist for the retainer’s tracking line:

• Prompt count frozen (write the number down)
• Owner named
• Cadence named (weekly / biweekly)
• Mention and citation are separate columns
• Search Console Web + gen-AI report if present, not instead of the panel
• Every red cell becomes a ticket or an explicit defer

If the agency cannot show last week’s table, they are not tracking. They are performing.

Why entity hygiene is not a one-week cleanup

Because campaigns, designers, and sales decks keep rewriting the facts. An audit can map the conflicts. A retainer keeps About, schema.org markup, footer, profiles, and the money page saying the same name, offer, and place. Models stitch those pages together. They do not pick your favorite Figma frame.

Fact	Where it usually drifts	What breaks in the answer
Legal / brand name	Homepage metaphor vs About vs schema name	Two companies, or a nickname the model treats as a different org
Offer sentence	Ads vs service page vs pricing	Cited for a package you do not sell
City / service area	Footer vs GBP vs location page vs old Webflow	The wrong town in the Overview
Price or “from” figure	Blog vs pricing vs PDF	Invented retainers — including yours
Founder / credential	LinkedIn vs About vs byline	E-E-A-T theater with mismatched years
Canonical domain	www vs apex vs shop subdomain	Citation to the zombie host

Cleanup procedure the audit starts and the retainer repeats:

1. Freeze one identity sentence and one offer sentence. Put them in a doc the designer cannot “improve” without a ticket.
2. Diff Home, About, money page, FAQ/proof, contact/location, JSON-LD, and the external profiles you actually control.
3. Ship the diffs in one deploy where you can. Staggered deploys are how you teach two cities for a month.
4. Re-run the identity prompts on the panel. File remaining inventions as corroboration or leftover third-party junk.
5. Next month, diff again. New campaign landing pages are the usual recidivists.

Checklist — entity line in the retainer:

• Single offer sentence in HTML on Home and the money URL
• Organization / LocalBusiness / Person markup matches visible text
• No leftover domain serving the old NAP
• Pricing claims either dated or removed from citeable passages
• Credentials you cite are ones you would repeat on a sales call (SEO since 2021 is mine; do not borrow it)

An audit that “fixes schema” once, then leaves the homepage in metaphor, is a half-cleanup. The retainer is there because someone will change the metaphor again.

What page-job work does the retainer ship that the audit only flags?

The audit labels protect, reinvent, or demote. The retainer changes the HTML. Informational pages that still exist to farm clicks will lose the argument with AI Overviews even if they rank. Money pages that read like glossaries will get quoted for a definition and skipped for the sale.

Page job	Audit output	Retainer output
Protect (pricing, demo, audit, high-intent service)	“Do not turn this into a blog”	Fast, accurate, unblocked; facts match the panel
Reinvent (what-is, how-to, comparison)	“Needs a citeable object”	Answer-first lead, table or procedure, handoff to a protectable URL
Demote	“Two quarters, no citation, no brand, no leads”	Merge, 301, or noindex with a written reason
Ignore this quarter	Explicit defer	Stays deferred — no stealth blog sprint

Rewrite skeleton the retainer actually ships (one URL at a time):

1. Lead answer in the first 60 words — the decision or definition, dated if it includes a number.
2. One citeable object — table, numbered procedure, or bounded checklist.
3. Entity facts that match About / schema.
4. A next step that is not “read more blog.”
5. Request indexing only after fetched HTML contains the answer.
6. Log the URL on the prompt panel for the queries it is supposed to win.

Do not ship as “page jobs”	Why
Twenty outlines, zero HTML	The audit already produced a list
New cluster to “feed the AI”	Volume is how you dilute the entity
Accordion-only FAQ	If it is not in the fetched HTML, it is not a source
nosnippet on the answer block	You opted out of the job
Homepage redesign as week one	Usually politics; rarely the citation

Two URLs shipped beats twelve in Notion. The retainer’s page-job KPI is shipped HTML on the ranked list, not “content in progress.”

What is citation recovery after the first loss?

A ticket, a page change, a corroboration check, and a re-measure — not a vibe that “AI is random.” You cannot recover a citation you never logged. That is why tracking sits in the retainer instead of in a quarterly audit.

Recovery procedure:

1. Confirm the loss on the same prompt, same surface, two captures if the first looks fluky. One miss is noise. A week of absence on a money prompt is a ticket.
2. Check eligibility: still indexed, still snippet-eligible, still the canonical you think it is. Use URL Inspection, not a sitemap assumption.
3. Check extractability: is the fact still in the first 60 words and in a table, or did a redesign bury it in a widget.
4. Check truth: did you change price, city, or package so the old cited sentence is now false. If yes, the model may have dropped you for being stale. Update the passage.
5. Check competitors: who got the slot. What object did they publish that you lack.
6. Ship the smallest page change that restores a citeable unit. Re-run the prompt for two more cycles before you declare failure.
7. If you never earned the citation, this is not recovery. It is the original page-job ticket from the audit. Do not relabel it.

Loss type	Likely cause	First move
Cited last month, absent now, page unchanged	SERP/answer churn or competitor object	Re-extract; add the missing table; do not panic-publish
Cited last month, absent now, you redesigned	Extractability or nosnippet	Restore the lead and the table in HTML
Mentioned, never cited	You are a named entity without a source URL the system trusts	Strengthen the URL and internal links; do not buy a mention dashboard
Cited with the wrong fact	Entity drift	Hygiene first; then the passage
Cited on a zombie domain	Canonical / migration	301 and Inspection; then panel

Google does not owe you a citation. Recovery is you remaining eligible, extractable, and true. The retainer is the cadence that notices the drop while the URL still has a chance.

What usually fails when teams buy the wrong object?

They buy a retainer as a second audit, or an audit as a citation SLA. Both fail in public, then someone declares “AEO does not work.”

Failure	What it costs	What you do instead
Retainer = monthly PDF of the same findings	Burn rate, no HTML, execs freeze the budget	Split SOW: audit once, then ship or staff
Audit = “we will get you cited this month”	You paid for a promise nobody can honestly make	Audit for baseline + ranked fixes only
Panel that changes prompts every run	Fake trends, no recovery tickets	Freeze 25–40, add only with a written reason
Rank tracker relabeled as AI visibility	You miss ChatGPT and Overviews	Panel + GSC; rank is a cousin metric
Page-job work that is only blog volume	Diluted entity, still no citeable table	Two reinvented URLs, then stop
Entity cleanup once, campaign undoes it	Wrong city in the answer for a quarter	Retainer hygiene or a freeze on rogue landing pages
Citation “recovery” with no prior log	You cannot tell loss from never-won	Audit’s prompt-panel v0 first
Success defined as old informational CTR	Every honest week looks like failure	Citation rate, brand search, protectable conversions

The expensive version is political: leadership bought a retainer to avoid staffing an owner. The agency cannot merge your schema. Your designer cannot freeze the offer sentence. Ninety days later everyone has screenshots and no shipped URL. That is not an AI problem. That is an ownership problem with an invoice.

In-house version of the same failure: you run a one-time audit, celebrate the deck, and put the tickets in a backlog behind a rebrand. The audit did its job. You used it as interior design.

How to read a vendor SOW before you sign:

Line in the SOW	Treat as	Ask this before you pay
“Monthly AI visibility report”	Report subscription until they name the four jobs	Where is the shipped-URL changelog?
“Includes strategy and implementation”	Mixed object — split it	What is the audit artifact vs the monthly artifact?
“Proprietary AEO platform access”	Tooling, not labor	Will you still get a CSV of the prompt panel if we leave?
“Ongoing optimization”	Empty	Name the URLs and the cadence
“We monitor ChatGPT weekly”	Tracking only	Who files the ticket when the answer is wrong?
“SEO retainer, now with AEO”	Classic SEO until proven	Show last month’s mention vs citation columns

If they cannot answer those in one call, you are not buying visibility work. You are buying a slide.

How do you measure whether the retainer is working?

With the frozen panel, Search Console, and pipeline on protectable URLs. Not with restored CTR. Not with a vendor letter grade. The dedicated method is measuring AI search visibility; this is the acceptance test for the contract.

KPI	Source	Review	Kill / change criterion
Prompt-panel citation rate	Manual log	Weekly	0% on money prompts for a quarter → change pages or stop
Mention accuracy (offer / geo / price)	Same log	Weekly	Repeated wrong facts after hygiene tickets → entity SOW is failing
Shipped ranked URLs	Git / CMS changelog	Weekly	Zero HTML in 30 days → you bought a report
GSC gen-AI impressions (if present)	Search Console	Weekly	Useful as reach, not as clicks; do not invent CTR from it
Brand queries	Search Console	Weekly	Free-fall after you deleted explainers → you cut demand
Protectable conversions	Analytics / CRM	Weekly	The number that pays rent
Eligibility regressions	URL Inspection	Weekly	New noindex / canonical loops → technical line is failing

Checklist at each monthly review:

• Same prompt set as month one (or a dated changelog if you added two)
• Citation and mention graphed separately
• List of URLs shipped, not “in design”
• List of recovered vs still-lost money prompts
• No slide that claims a return-on-ad-spend for citations
• No slide that quotes a viral Overview CTR study as your result

As of August 2026, public CTR studies on AI Overviews disagree on magnitude. They are not a retainer KPI. Quote them only as context. Your panel and your Search Console are the decision.

If informational sessions fall while citation rate, brand search, and protectable conversions hold, the retainer may still be doing the job. If all four fall, you have a demand problem or you bought screenshots.

Monthly review agenda (30 minutes, one page):

1. Prompt-panel deltas: gained citation, lost citation, still absent, new wrong fact.
2. Shipped HTML: URLs, date live, Inspection status.
3. Entity diffs: what a campaign tried to change; what you froze.
4. Recovery tickets: open, shipped, abandoned with a reason.
5. Pipeline on protectable URLs — not blog sessions.
6. Decision: continue, cut a workstream, or end the SOW.

Review outcome	Evidence	Next month
Keep	Panel owned; ≥1 URL shipped; facts more accurate	Same four jobs
Cut tracking theater	Screenshots, no tickets	Fire the report line; keep shipping
Cut page-job fiction	Outlines only	Pause retainer until a developer week exists
End	Zero HTML and a rotating prompt set	You have a PDF. Stop paying for it

What should you sequence in 30, 60, and 90 days?

Audit in the first stretch if you lack a baseline. Retainer workstreams after the list is real. Do not staff a year on day two. Do not skip the week-one eligibility check because it is boring.

Window	If you are starting from zero	If the audit already exists
Days 1–10	Eligibility + Search Console + prompt-panel v0	Re-run the panel; throw out stale tickets
Days 11–30	Audit readout: baseline + prioritized fixes; ship zero or one obvious technical fix	Ship two ranked URLs (page jobs); freeze entity sentences
Days 31–60	Decide: in-house vs retainer vs stop	Weekly panel; entity diffs; first recovery tickets
Days 61–90	Only buy a retainer if HTML shipped and the panel is still owned	Show citation / mention trend + shipped changelog, or kill the SOW

If you only have a week, skip the retainer conversation entirely.

Week-only checklist:

• Skip: buying a six-month retainer, a new blog calendar, llms.txt theater, AI-only schema
• Do: URL Inspection on the homepage and one money URL
• Do: run 15 buyer prompts once, by hand, and write mentioned / cited / wrong
• Do: freeze the identity sentence and the offer sentence
• Do: book an audit if the answers shocked you, or staff two URL rewrites if they did not
• Stop at the end of the week. Do not “start the retainer” with no baseline

Ninety days with a retainer and no shipped URL is not a slow start. It is the wrong object. End it.

Staffing decision after the audit, before a retainer invoice:

You have	Next object	Do not
Developer + operator who can ship two URLs a month	Keep the work in-house; re-audit in a quarter if the panel dies	Pay for screenshots of work you can do
Operator, no developer	A sprint or developer week, then decide	A visibility retainer that cannot merge schema
Developer, no operator	Audit already failed — nobody owns the sentences	Retain an agency to “be the founder”
Neither, messy eligibility	Eligibility fix, then audit	A six-month AEO package
Both, list idle for 90 days	Retainer on the four jobs, with a 90-day kill switch	An open-ended “optimization” SOW

Procedure for the kill switch, written into the contract:

1. Month 1 must include the frozen prompt file and at least one shipped ranked URL, or the SOW converts to a wind-down.
2. Month 2 must show mention and citation as separate columns, not a blended “visibility score.”
3. Month 3 must show either a recovered money prompt, a documented reason it is still absent, or a page-job change aimed at that prompt.
4. If those three fail, you end. You do not “give it another quarter” because the dashboard looks expensive.

When is this not worth doing yet?

When the site cannot be a source, the offer is fiction, or nobody will own the panel. Paying for either object in that state buys a PDF about a wall.

Blocker	Why audit and retainer both stall	What to do first
Sitewide noindex, login wall, or nosnippet on answers	Not snippet-eligible	Remove the accident; read AI features
Domain not in Search Console	You cannot measure	Verify the property
Legal name in flux	You will teach two entities	Freeze the name, then write
No offer you will take next month	Money page is fiction	Pause marketing pages
Duplicate live hosts (old site + new site)	Models cite the zombie	301 the zombie
No one can edit HTML	Schema and leads will stay wrong	Get a developer week or a sprint, not a visibility retainer
Success must be “old CTR back”	You will fire the honest report	Change the scoreboard first
You want a citation SLA	Nobody honest will sign it	Buy baseline + shipping, not a guarantee

Proceed only if:

• You can name the offer you will take
• Canonical domain is clear
• Search Console exists or will this week
• Someone can change server-rendered HTML
• You will accept mention / citation / conversions as the scoreboard
• You will not demand a monthly fee quoted from a blog post (this one will not give you one)

If those boxes are empty, fix the business constraint. Then an audit. Then a retainer only if the list would otherwise rot. The booking path for the diagnosis is a visibility audit. The system map is the AEO playbook.

Bravery is not a citation SLA.

FAQ

What belongs in an AI visibility monthly retainer vs a one-time audit?

A one-time audit is the baseline plus a prioritized fix list: eligibility, a prompt-panel v0, an entity map, protect-versus-reinvent labels, and a ranked 30/60/90. A monthly retainer is prompt-panel tracking, entity hygiene, page-job rewrites, and citation recovery — the labor that repeats after the readout. If the monthly artifact is the same PDF, you bought a report subscription.

How do I measure whether an AI visibility retainer is working?

Score a frozen prompt panel (mention vs citation vs accuracy), count shipped URLs on the ranked list, and watch brand search plus conversions on protectable URLs. Use Search Console generative-AI impressions if the property has them; they are not clicks and they are not ChatGPT. Do not use restored informational CTR or a vendor letter grade as the acceptance test.

What usually fails first when teams try this?

They buy a retainer before they have a baseline, or they treat the audit as a citation guarantee. The next failure is a prompt set that changes every week, so nothing can trend, and a month of screenshots with zero HTML. Ownership fails before the model does.

How long does this take to show results?

Eligibility issues can show up in days once Inspection is clean. Panel accuracy often moves in weeks after entity and lead-answer fixes. Citations are not guaranteed and are not on a published SLA — Google’s docs treat indexing and serving as never-promised. Re-measure on the same prompts at two weeks and eight weeks; do not invent a day-count.

What should I skip if I only have a week?

Skip buying a retainer, skip a new blog calendar, and skip AI-only markup. Spend the week on Inspection, 15 handwritten prompts, and freezing the identity and offer sentences. Book an audit if the answers shocked you, or staff two URL rewrites if they did not — then stop.

When is this not worth doing yet?

When the site is not snippet-eligible, Search Console is missing, the legal name is mid-rebrand, or you do not have an offer you will take. Also skip it if success must be “get our old CTR back” or a signed citation guarantee. Fix eligibility and the scoreboard first; then buy a baseline, then buy shipping.

CTA

If you cannot name the baseline, buy the audit. If you can name it and nobody is shipping, buy the four retainer jobs — not a monthly reprint of the same PDF.

Lane: /visibility · Book a visibility audit.