Skip to main content Scroll Top

How to Get Cited by AI: The Nine Moves, In Order

Updated September 2026 · Written and maintained by the Progression Agency strategy team

Three of them are free and take an afternoon. One dominates the budget. The order matters more than the list, because everything after move four assumes the machine can read you.

On this page · 13 sections
  1. How to get cited by AI
  2. The three free moves, done today
  3. The expensive move: rendering
  4. The highest-ratio content change there is
  5. Record a baseline before you change anything
  6. The slow half: corroboration
  7. What not to do
  8. How fast should you expect this to work?
  9. Everything else we have written on search, AI and getting found
  10. The nine moves as a work order
  11. Passage rewriting, before and after
  12. Advice that does not work, and why
  13. Who gets there fastest, and who should wait

The short answerNine moves, in order: allow the five AI crawlers, prove from your server logs that they arrive, fix status codes your CDN or WAF returns to them, render your content without JavaScript, move each answer to the first sentence of its section, source claims inline, unify your entity data, record a dated prompt baseline before anything changes, and build third-party corroboration. Moves one to three cost nothing and take about an hour. Move four decides your budget. Move nine is slow and is usually why a competitor is cited and you are not.

Time-to-signal figures are our own observation across audits, labelled as such. Nobody controls attribution, so nothing here is a guarantee of citation.

The order of operationsThe order of operations
Steps one to three cost almost nothing and take an afternoon. Step four is the budget decision. Steps five to nine are ordinary discipline pointed at a new target.

How to get cited by AI

Nine moves, in order. Three of them cost essentially nothing and can be done this afternoon. One dominates the budget. The remaining five are ordinary discipline pointed at a new target.

The order matters more than the list. Every step after the fourth assumes the machine can actually reach and read your page, and a large share of the effort spent on answer engine optimization is spent on pages that fail that assumption silently.

Start at step one even if you are confident. We have found blocked crawlers on sites whose owners were certain nothing was blocked, because the block came from a security plugin default rather than from anything anyone chose.

Step 1 — Allow crawlers. Free. Minutes. Binary.
Step 2 — Prove arrival. Server logs, not assumptions.
Step 3 — Fix status codes. CDN and WAF, not just robots.txt.
Step 4 — Render without JS. The budget decision.
Step 5 — Answer in sentence one. Cheap, sitewide, helps readers.
Step 6 — Source inline. Survives reliability filtering.
Step 7 — Unify entity. One description, everywhere.
Step 8 — Record baseline. Before you change anything.
Step 9 — Corroboration. Slow, durable, decisive.
Ongoing — Re-run the prompt set. Identical wording, monthly.
Ongoing — Watch the logs. Crawler frequency is a live signal.
Ongoing — Report separately. Never merge with rankings.

The three free moves, done today

Read your robots.txt for the five AI crawlers. Request a page with each of their user agent strings and confirm a 200 rather than a 403. Then grep your server logs to see whether they have actually arrived in the last thirty days.

Together these take about an hour and they are the difference between being a candidate for citation and being entirely absent. On a meaningful minority of sites, this hour is the whole result for the quarter.

robots.txt — Quick win. A Disallow you did not know about.
403 to bots — Quick win. A WAF rule nobody remembers adding.
Answer position — Quick win. Move it to sentence one.
Missing sources — Quick win. Attach them inline.
Stale description — Quick win. One canonical version.
No baseline — Quick win. An hour of work, permanently useful.

Move one: allow the crawlers

GPTBot, OAI-SearchBot, ClaudeBot, PerplexityBot and Google-Extended. A Disallow under any of them is a complete explanation for absence on that surface, and removing it costs nothing.

Move two: prove they arrive

Permission is not evidence. Filter your access logs for those user agents. If robots.txt permits them and the logs show nothing, something else is blocking.

Move three: fix the status codes

This is the one people miss. Your CDN or WAF may return a 403 to unrecognised agents while serving browsers perfectly. It is invisible unless you test with the user agent string set.

The expensive move: rendering

Fetch a commercial page with JavaScript disabled. If what comes back is a navigation bar, a footer and little else, the model has never seen your content — and fixing that is an engineering project rather than a marketing task.

This single test decides the size of everything else. A server-rendered site can complete most of this plan in a few weeks. A client-rendered site is looking at a rendering programme first, and no amount of content work substitutes for it.

Effort against time-to-signalEffort against time-to-signal
The first two produce checkable evidence within days. Nothing after step five produces an honest reading in under a month, which is why the baseline in step eight matters.

How to test it

Disable JavaScript in your browser’s developer tools and reload, or fetch the URL with curl and read what comes back. Compare the word count against the rendered page.

What good looks like

Your actual body content present in the HTML before any script runs.

What to do if it fails

Server-side rendering or prerendering for the commercial templates. Not the whole site necessarily — the templates that carry the pages you want cited.

Why it is worth the cost

Every other step on this page is conditional on it. Structure, sourcing and entity work applied to a page that returns nothing are applied to nothing.

The highest-ratio content change there is

Move the answer to the first sentence of the section that promises it. That is the whole change, it costs almost nothing, it applies to every page you own, and it improves the page for human readers at the same time.

Retrieval selects a passage rather than a page. A section whose heading asks a question and whose fourth paragraph answers it gives a system nothing clean to lift, and a thinner competitor with the answer up front gets quoted instead.

How to rewrite a section so it gets quotedHow to rewrite a section so it gets quoted
The test at the end is the whole method: a passage that survives being read out of context is a passage that can be lifted into an answer.

Make the heading a question

Or at least a clear promise. A heading that says ‘Our approach’ promises nothing a retrieval system can match against.

Answer it immediately

Complete sentence, no preamble, no throat-clearing. The answer should survive being read entirely alone.

Context goes after

Human readers cope with context-then-answer. Retrieval does not.

One idea per section

Two answers in one section usually means neither extracts cleanly.

Attach sources inline

An unsourced statistic is among the first things a cautious system declines to repeat.

The final test

Read the passage on its own, out of context. If it still makes sense, it can be quoted. If it needs the paragraph above it, it cannot.

Record a baseline before you change anything

Thirty buyer questions, asked across two or three assistants, with every answer stored verbatim and dated. An hour of work, and it is the difference between being able to prove a result later and merely asserting one.

Do it in week one, before any of the fixes land. A baseline recorded after the work has started makes every subsequent claim unfalsifiable, including your own honest ones.

A realistic first eight weeksA realistic first eight weeks
Note the baseline lands in week one, before anything changes. Recording it later makes every later claim unfalsifiable.
Crawler hits — Evidence. Server logs, within days.
Render coverage — Evidence. Scriptless word count per template.
Passage extraction — Evidence. Read the section standalone.
Entity agreement — Evidence. List every source, compare.
Corroboration — Evidence. The actual third-party pages.
Citation frequency — Evidence. Fixed prompt set, stored answers.

The slow half: corroboration

Independent third parties saying about you what you say about yourself. It is the slowest thing on this list, the hardest to fake, and in our experience the usual reason a competitor is cited and you are not once the technical work is done on both sides.

There is no shortcut here that survives contact with a retrieval system. Digital PR, original data worth citing, genuine expert contribution, and presence in the listicles and directories that already rank for your category.

Where effort goes, month one versus month sixWhere effort goes, month one versus month six
The inversion is the point. Month one is technical and cheap; month six is corroboration, which is slow, expensive and the thing that actually separates you from competitors long-term.

Original data

The strongest form, because it makes you the primary source rather than a commentator. Anything you measured that nobody else publishes qualifies.

Expert contribution

Being quoted in other people’s work. Slow to build, disproportionately effective, and largely a matter of being responsive and useful.

Category listicles

‘Best X agency’ answers come from lists rather than from agency pages. Getting onto those lists is a different activity from optimising your own site.

Directories that rank

Clutch, G2 and their equivalents are retrieved heavily for provider queries. Presence there is corroboration whether or not anyone clicks through.

Consistency underneath it all

Corroboration only helps if the third parties describe you the same way you describe yourself. This is why entity work precedes PR work.

What not to do

Six pieces of advice you will be given that are either neutral or actively counterproductive: publish more content, increase keyword density, add llms.txt and wait, buy a citation guarantee, measure it with a rank tracker, and pad your FAQs with questions nobody asks.

Each of these is currently being sold. Volume and density do not move retrieval because similarity is computed on meaning. Rank tracking cannot see inside an answer. And nobody controls attribution, so nobody can guarantee it.

Advice you will be given that does not workAdvice you will be given that does not work
Six reds. Each one is currently being sold by somebody, and each one is either neutral or actively counterproductive.
More articles — Not a lever. Volume does not drive retrieval.
Keyword density — Not a lever. Similarity is on meaning.
llms.txt alone — Not a lever. Low adoption, not a ranking factor.
Rank tracking — Not a lever. Cannot see inside an answer.
Guarantees — Not a lever. Attribution is not addressable.
FAQ padding — Not a lever. Questions nobody asks retrieve nothing.

How fast should you expect this to work?

Crawler access shows in your logs within days. Rendering fixes land in a week or two once engineering ships them. Passage restructuring improves extraction over weeks. Citation frequency does not read honestly before about month six.

The useful early signal is not a citation. It is a crawler hit, then a scriptless fetch that returns your content. Those two are checkable, fast, and they tell you the chain is intact.

Worth doing, and not worth doingWorth doing, and not worth doing
The three reds are the three things most commonly sold. Volume and density do not move retrieval, and nobody controls attribution.
Prioritising your nine movesPrioritising your nine moves
Top-right first, always. Bottom-right — rendering and corroboration — is where the budget actually goes, and both are worth it once the free wins are banked.
The shape of the workThe shape of the work
Three free steps and one expensive one. Establishing which category you are in is the entire first week.
Who wins fastest — Server-rendered sites. Skip the expensive step entirely.
Who wins fastest — Few templates. One fix applies everywhere.
Who wins fastest — Question-led categories. Buyers already ask assistants.
Who waits longer — App-like sites. Rendering is an engineering project.
Who waits longer — New domains. Corroboration takes time to accumulate.
Who should not bother yet — Local trades. Listings still return more today.

Want the first week done properly?

An AI visibility audit runs moves one through four, reads your server logs, and records the baseline before anything changes — so the rest of the plan can actually be proven later.

/ai-visibility-audit

Want this done for your site?We build and maintain the search, content and paid programmes described on this page.

Get a free proposal

Everything else we have written on search, AI and getting found

AI, AEO and what is changing

Websites and design

Choosing and working with an agency

Social, content and brand

By industry and by situation

Not sure which of these applies to you?Tell us the situation and we will say plainly what we would do first, and what we would not.

Talk it through

The nine moves as a work order

The nine moves in the order we would do them, with the effort and the evidence each produces.

Effort, cost and time to a checkable signal
#MoveEffortCostTime to signalHow you verify it
1Allow the five AI crawlersMinutesFree2 daysServer logs show the agents
2Prove they arriveMinutesFree2 daysLog entries with 200 status
3Fix status codes from CDN/WAFHoursLow3 dayscurl with the user agent set
4Render without JavaScriptWeeksHigh~30 daysScriptless word count per template
5Move answers to sentence oneDaysLow~21 daysRead the passage standalone
6Source claims inlineDaysLowWeeksEvery statistic has attribution
7Unify entity dataWeeksMedium~75 daysEvery source describes you identically
8Record a prompt baseline1 hourFreeImmediateDated file of verbatim answers
9Build corroborationOngoingHigh~150 daysThird-party pages that mention you

Passage rewriting, before and after

The same passage in its original and rewritten forms, with what changed and why it matters.

The same information, structured two ways
BeforeAfter
HeadingOur approach to pricingHow much does an AEO audit cost?
First sentenceWe believe transparency matters to our clients.A published AEO audit costs $1,000 to $4,000 for 10 to 20 hours of work.
Second sentenceEvery business is different, and so is every engagement.The range depends on template count and rendering health.
ContextArrives in sentence oneArrives after the answer
Extractable alone?No — needs the paragraph below itYes — survives out of context
Likely outcomeRanks fine, rarely quotedQuotable as a standalone passage

Advice that does not work, and why

Common recommendations in this field that we would not follow, with the reasoning.

Six common recommendations, assessed
AdviceVerdictWhy
Publish more articlesDoes not helpRetrieval selects passages, not volume
Increase keyword densityDoes not helpSimilarity is computed on meaning, not term frequency
Add llms.txt and waitMarginalLow adoption, not a ranking factor. Hygiene at best
Buy guaranteed citationsNot possibleAttribution happens inside the model; nobody controls it
Track with a rank trackerActively misleadingCannot see inside an answer; reports success while absent
Pad FAQs with extra questionsDoes not helpQuestions nobody asks retrieve against nothing

Who gets there fastest, and who should wait

The situations where this work pays quickly, and the ones where it should be deferred.

Realistic expectations by situation
SituationTime to meaningful movementWhy
Server-rendered site, few templatesWeeksSkips the expensive step entirely
Server-rendered, many templates1-2 monthsStructure work scales with template count
App-like site, in-house developers2-3 monthsRendering is a project but you can ship it
App-like site, no developersBlocked until resourcedContent work on unreadable pages returns nothing
New domain, any stack6+ monthsCorroboration has not accumulated yet
Local trade or walk-in retailNot yet worth itListings and reviews still return more today

Getting found in search

AI, AEO and what is changing

Paid media and lead generation

Websites and design

Choosing and working with an agency

Social, content and brand

By industry and by situation

Frequently asked questions

How do I get cited by AI?
Nine moves in order: allow the five AI crawlers, prove from server logs that they arrive, fix status codes returned by your CDN or WAF, render your content without JavaScript, move each answer to the first sentence of its section, source claims inline, unify your entity data, record a dated prompt baseline, and build third-party corroboration.
What is the first thing I should do?
Read your robots.txt for GPTBot, OAI-SearchBot, ClaudeBot, PerplexityBot and Google-Extended. It takes two minutes, costs nothing, and on a meaningful minority of sites it is the entire explanation for absence.
How long does it take to get cited?
Crawler access shows in logs within days. Rendering fixes land in a week or two. Passage restructuring improves extraction over weeks. Citation frequency does not read honestly before about month six.
What is the single highest-impact content change?
Moving the answer to the first sentence of the section whose heading promises it. Cheap, applies to every page, and it improves the page for human readers at the same time.
How do I know if my pages are readable by AI crawlers?
Fetch a commercial page with JavaScript disabled, using curl or your browser’s developer tools. If what returns is a navigation bar and a footer, the model has never seen your content.
Does publishing more content help?
No. Retrieval selects passages rather than rewarding volume. One page whose section answers the question cleanly beats twenty that circle it.
Does keyword density matter?
No. Similarity is computed on meaning, so repeating a phrase does not increase your odds of being retrieved.
Should I add llms.txt?
You can, but do not expect it to change anything. Adoption is low and it is not a ranking factor. Treat it as hygiene, not strategy.
Can I buy a guaranteed citation?
No. Attribution happens inside the model, there is no ad inventory and no submission process. Any supplier offering a guarantee is describing something that does not exist.
Why record a baseline before making changes?
Because answers vary between runs. Without a dated record of verbatim answers from before the work started, no later claim about improvement can be verified — including an honest one.
What goes in a prompt baseline?
About thirty questions in your buyers’ real language, asked across two or three assistants, with every answer stored verbatim, dated and labelled by platform.
What is corroboration and why does it matter most?
Independent third parties saying about you what you say about yourself. It is the reliability signal a retrieval system can actually use, and once technical work is done on both sides it is usually why a competitor is cited and you are not.
How do I build corroboration?
Original data worth citing, expert contribution to other people’s work, presence in the category listicles that already rank, and directory profiles like Clutch and G2 that get retrieved for provider queries.
Why does my CDN matter?
Because bot protection frequently returns a 403 to unrecognised user agents while serving browsers perfectly. It is invisible unless you test with the crawler user agent string set.
Should I fix rendering or write content first?
Rendering. Structure, sourcing and entity work applied to a page that returns nothing to a crawler are applied to nothing.
Can I measure this with a rank tracker?
No, and doing so is actively misleading. A rank tracker cannot see inside an answer and will report steady success while you are absent from every answer a buyer receives.
Which platform will cite me first?
Usually Perplexity, because it fetches live at question time and depends far less on deep indexation than an index-led surface.
How many prompts should I track?
About thirty to start. That is manageable by hand and large enough to show a trend when re-run monthly with identical wording.
What if I do all nine moves and still am not cited?
Then you are likely competing on corroboration against a better-established competitor, which is the slow problem rather than the broken one. That is a digital PR and original-data question.
Is this worth doing for a local business?
Usually not yet. For local trades and walk-in retail, listings, reviews and an accurate Google Business Profile still return more per pound than this work does today.
Do I need an agency for this?
The first three moves you can do yourself in an hour. Rendering usually needs developers. The rest is discipline. An agency is worth it mainly for the diagnosis and for recording the baseline properly.
What should I check monthly?
Crawler hits in your logs, scriptless render coverage on any new templates, and the prompt set re-run with identical wording — reported separately from search rankings.

Want this done for your site?We build and maintain the search, content and paid programmes described on this page.

Get a free proposal

Get a free marketing proposal

Tell us what you are trying to grow and we will come back with a plan, not a pitch deck. Same-day reply on weekdays.

Our website uses cookies from third party services to improve your browsing experience. Read more about this and how you can control cookies by clicking "Privacy Preferences".
Privacy Preferences
When you visit our website, it may store information through your browser from specific services, usually in form of cookies. Here you can change your privacy preferences. Please note that blocking some types of cookies may impact your experience on our website and the services we offer.
Privacy Policy
You have read and agreed to our privacy policy
Required
Contact Us