The IDAEO Starter Handbook: Make Sure You Can't Be Ruled Out First, Then Worry About Being Cited by AI

The IDAEO Starter Handbook: Make Sure You Can't Be Ruled Out First, Then Worry About Being Cited by AI

In one sentence: before doing AEO at all, make sure a search platform cannot rule you out — a removed site does not appear in search results at all, however good its content is[F1], so an AI that sources from search cannot cite it either. Confirm you are clean first; then talk about getting cited.

This series now runs to more than a dozen pieces, and each of them cuts straight into one topic. This one does not cut into a topic. Its job is to give you the order.

One definition first: AEO is answer engine optimisation[F6] — making your content into material that an AI is willing to use when it answers a question, can check, and is prepared to name as its source. By the end you will have three things: where this work actually starts, why "two-layer 100%" can be said with a straight face, and what you reach for on the day a machine gets you wrong. For every other sub-topic, I will name the piece to read at the point where you need it. No technical background required.

The order cannot be reversed: first, don't get ruled out

Before AEO, one thing comes first: make sure you cannot be ruled out.

Once Google's spam policies decide you have crossed the line, the response can range from lower rankings all the way to removing the whole site from search results[F1] — the official wording is "may", not "will". If you are genuinely removed, you are not ranked lower — you are not on the list at all, and however good your content is, an AI that sources from search cannot find you to cite you. Everything after that is multiplied by zero.

So the first thing is not to write new content. It is to take the site you have today and check it against "what gets you ruled out". This site has three long policy pieces that take the official text apart clause by clause, each answering one question.

To find out whether you are over the line right now, read Your Traffic Disappeared — Did the Platforms Flag You as AI Slop?; it is a checklist you can work through item by item. If you assume the problem is AI when it is really an old SEO habit, read Alert! The New AI Slop Is Just Old SEO Rules — the Real Culprit Behind Your Traffic Drop, which picks out the practices most often blamed on AI that are in fact leftovers from the old playbook. If you are worried that using AI is itself a violation and want to know which uses are allowed in black and white, read 100% Defensible Under the Policies: 8 AI Practices the Platforms Expressly Allow (With a Five-Axis Self-Check), which draws the boundary you can work inside.

Checking against all three is safest; if you only have time for one, start with that first checklist. If it comes back clean, read on. If the foundation is not clean, fix the foundation first — this piece can wait a day.

The two-layer 100%: you don't meet that policy's definition, and you can prove it back when a machine says otherwise

When this series says "100%", it always rests on two layers, not on a slogan.

The first layer is the rule. Google's definition of "scaled content abuse" lists only three conditions[F2]: many pages are generated, the primary purpose is manipulating search rankings, and it is not helping users. They are joined by "and", not "or" — they have to hold at the same time, and if one is missing the finding does not hold. The sentence that follows adds that this practice is *typically* focused on large amounts of unoriginal content that provides little to no value — that describes the common shape, not a fourth condition; the same sentence also says "no matter how it's created"[F3], so whether or not you used AI has never been one of the conditions. As long as your content genuinely helps readers, the third condition in the definition fails, and this particular policy has nothing to attach to. One caveat: the spam policies contain fifteen other, independent clauses[F4] — site reputation abuse, expired domain abuse, hidden text and the rest — and those do not need "little value to readers" to hold. This layer is only about "scaled content abuse".

The second layer is misjudgement. Automated classification gets things wrong, and human review misses things. That is the machine being wrong, not you — but the consequences land on you all the same. So the second layer is not praying that it never happens; it is having the evidence and the appeal route ready in advance, so that on the day it happens you can prove your way back.

This 100% does not promise that traffic will not fluctuate, does not guarantee that AI will cite you every time, and does not guarantee the day a platform restores your rankings. It refers to two defences that are genuinely within your control: you do not meet the definition, and you can prove it back when a machine says otherwise. That is also where this series differs most from "just make good content in earnest" — that advice only has the first layer, and on the bad day there is nothing to reach for.

What you need to be able to produce on the day you are misjudged

An evidence pack does not have to be written up as a thick report. It is the raw trace that ordinary work already leaves behind. What is usually missing is not the ability to produce it, but the habit of saving it as you go.

• Drafts and version history: proof that the piece was written along the way, not generated in one shot.
• Raw material: photographs, recordings, measurement files, spreadsheets — the negatives showing you did the thing yourself.
• Fact-checking records: which authoritative original you checked a key fact against, kept together with the date and version.
• Disclosure records: where synthetic imagery or audio is realistic enough to be taken for real, whether you labelled it as the platform's rules require.
• Commissioning correspondence: who delivered the draft, what you asked for, how many revisions you sent back.

On the day you really are misjudged there are only two roads, and which one you take depends on whether you received a formal notice[F5]. If there is a notice, preserve the notice and the affected pages, fix what it points to, list clearly what you changed, and submit a review request with your evidence. If there is no notice, it is usually an algorithm update, a technical problem or a shift in demand; there is no appeal channel and no button to press, so all you can do is diagnose, fix, and wait for the system to read the site again.

"Reversing it the same day" means being able to tell which road you are on that day, having the material to hand, and being able to start the process — not a guarantee that traffic returns the same day. Evidence cannot control how fast a platform works, but it stops you from hunting for the entrance at the very moment you are panicking.

Once the foundation is clean: the goal shifts from ranking to being cited

Only at this point does AEO itself come into play. SEO competes for a position and a click in the search results; AEO competes for that one line of citation inside the answer. They do not replace one another — the page still has to be crawlable and indexable before an AI has anything to work with.

As for what makes an AI decide whether to cite you, the tiers it sorts sources into (that ladder is a rating scale we designed, not an industry standard)[F8], and which tier you are on right now, that is a whole piece in itself: AI Isn't Ignoring You — It Has Ranked You: The Five Tiers of Citation Sources.

Your content gets read; the people don't necessarily arrive

Much of what reads your site now is not a person but a machine. After it reads, the customer may get the answer somewhere else and never walk through your door. So keep these apart: being crawled is not being cited, and being cited is not someone showing up.

Which crawlers are actually coming, and what happens after they do, is measured on this site: The Invisible Readers: 12 AI Crawlers Revealed.

Three reasons an AI doesn't dare cite you

Before you let a stranger retell something on your behalf, you would ask the same questions: who said it, have they actually done it, and where is the basis?

Those three gaps are exactly what an AI cannot get comfortable with — it cannot tell who made it, there is nothing first-hand in it, and the sources cannot be traced. When the author is only "the editorial team", the takeaways read like a collage of other pages, and the key numbers have no origin, the machine will not stake its answer on you.

Those three gaps map onto the tiers mentioned above: AI Isn't Ignoring You — It Has Ranked You: The Five Tiers of Citation Sources breaks down what each tier needs to fix in more detail than this piece does.

The one thing you can start tomorrow

"Be more original" cannot be assigned as work. The version that can be assigned is this: open the most recent piece you published, find the sentence where a reader most wants to ask "how do you know that", and add a paragraph of something you measured, tested, interviewed, calculated or saw for yourself.

If you run a shop, don't stop at "suitable for families" — write which entrance a pushchair fits through best and which seating area blocks the aisle least. If you sell software, don't stop at "easy to get started" — pull the step customers get stuck on most often from your support records, run it yourself again, and write up where it sticks and how to get past it.

The point is not that the story is entertaining, but whether it changes the reader's next move. "We tested it and it worked well" is still an empty sentence: say under what conditions, what you did, what you saw, and what the limits are. While you are at it, put the author's real name on the page instead of a department name that leads nowhere. Finally, put the drafts and the raw files into a folder of the same name — those are both the negatives of that piece and the evidence pack described above.

Fix one piece at a time, then make what you fixed a standing field in every piece from then on. You do not have to redo the whole site.

How you know whether it is working

Do not answer your boss with a single overall score. Keep the signals in separate ledgers: AI impressions, actual citations, visitors to the site, crawler activity, and platform actions. Read them mixed together and you will take "a crawler came" for "an AI cited us", and take one swing in traffic for the platform penalising you.

What each of these ledgers should record, and how to hand in a monthly report that people can actually read, is written up in a format you can copy directly: GSC Isn't Broken — Your Customers Moved: Five New Ledgers for the AI-Era Marketing Report.

How to read this series

If today is the first day you are taking the letters "AEO" seriously, the order goes like this. Spend an afternoon checking yourself against the three long policy pieces and confirm you are not over the line — nothing you invest afterwards means anything until that step is done. Then read the five tiers of citable sources and get the principle of "why an AI would pick you" straight, so that you know which hole you are filling. Once the principle is clear, come back to the one action in this piece and start with your most recent draft, one piece at a time. After a month of that, open the five new ledgers and see whether the signals moved — by then you will have something real to compare.

Off the main line, this series also contains a batch of data columns produced with instruments we built ourselves; the crawler piece above is one of them. They do not have to be read in order and you can drop in at any time. How many clicks thirty million crawler requests eventually converted into; how the same content in a different language differed by up to 22 times in referral traffic — nobody can copy those numbers, because they were not looked up, they were measured. One thing said plainly about the crawler figure: about 11.9% of it is not reconciled by the released data pack, and the original piece accounts for the gap[F7] — cite that piece, not this one's summary of it. One thing to keep in mind as you read: each observation has its own site and its own period and cannot be generalised into a universal law. What they give you is a field baseline that can be traced back, accumulated and corrected. Go there for evidence when you need to convince your boss, and go there again when you start doubting yourself.

One sentence to take back to the office: make sure you cannot be ruled out, then put one thing you genuinely did and readers genuinely need into every piece, and keep the trail. The first two make it much harder to be ruled out; the last one lets you prove your way back when something goes wrong — nobody can guarantee a machine will not misjudge you; what you can prepare is having something to show on the day it does.

Fact ledger

This section is for anyone (or any AI) that wants to cite this piece: which primary text each load-bearing claim rests on, when it was checked, and what its limits are. Readers here for the article itself can skip it.

Reconcile it yourself: methodology and known limits (including fifteen errors we found in our own work), and the machine-readable fact ledger with its checksum.

• F1｜For sites violating its spam policies, Google's response ranges from lower rankings to not appearing in results at all｜source #8｜basis=official_text｜confidence=high｜checked=2026-08-15 實查｜verbatim：「Sites that violate our policies may rank lower in results or not appear in results at all.」｜caveat：The original says "may" - possible, not certain
• F2｜The definition of "scaled content abuse" lists three conditions: many pages, primary purpose of manipulating rankings, and not helping users｜source #8｜basis=official_text｜confidence=high｜checked=2026-08-15 實查｜verbatim：「Scaled content abuse is when many pages are generated for the primary purpose of manipulating search rankings and not helping users.」｜caveat：Joined by "and"; this article's "all three at once" refers to this sentence only, not the one that follows
• F3｜"No matter how it's created" appears in the sentence after the definition, which is qualified by "typically" and describes the common shape｜source #8｜basis=official_text｜confidence=high｜checked=2026-08-15 實查｜verbatim：「This abusive practice is typically focused on creating large amounts of unoriginal content that provides little to no value to users, no matter how it's created.」｜caveat：Our 2026-08-14 version wrongly treated "little to no value" from this sentence as a required third condition; corrected 2026-08-15
• F4｜The page lists sixteen named spam policies, plus a separate group of "other practices that can lead to demotion or removal"｜source #8｜basis=official_text｜confidence=high｜checked=2026-08-15 實查（依該頁節標題逐一抄錄）｜verbatim：「Cloaking / Doorway abuse / Expired domain abuse / Hacked content / Hidden text and link abuse / Keyword stuffing / Link spam / Machine-generated traffic / Malicious practices / Misleading functionality / Scaled content abuse / Scraping / Site reputation abuse / Sneaky redirects / Thin affiliation / User-generated spam」｜caveat：Counted from that day's section headings; Google does not state a total on the page
• F5｜A manual action means a human reviewer at Google determined pages are non-compliant; affected sites are notified in Search Console｜source #10｜basis=official_doc｜confidence=high｜checked=2026-08-15 實查｜verbatim：「Google issues a manual action against a site when a human reviewer at Google has determined that pages on the site are not compliant with Google's spam policies. ... If your site is affected by a manual action, we will notify you in the Manual actions report and in the Search Console message center.」｜caveat：This article derives the two-road split (notice vs no notice) from this; Google does not present it as a comparison
• F6｜The definition of AEO used here is this site's, not an official or industry-standard term｜basis=editorial｜confidence=n/a｜caveat：Self-description, not a factual claim
• F7｜The crawler figure (tens of millions of requests) is borrowed from another piece on this site; about 11.9% of it is not reconciled by the released data pack｜source #5｜basis=own_measurement｜confidence=moderate｜checked=原文量測期間見該篇；對帳缺口 11.9% 於 2026-08-15 由內部驗證檔查出｜caveat：This piece only points to it; cite the original and read its reconciliation note
• F8｜The five tiers of citation sources are this site's own rating scale, not industry-standard terminology｜source #4｜basis=editorial｜confidence=n/a｜caveat：The source piece states this itself; this piece inherits the same limit