Citation durability: the 90-day number AEO specialists should publish

A citation counts when it is still there a quarter later.

Ask ChatGPT, Perplexity or Google AI Overviews which companies specialize in answer engine optimization, and the reply arrives as a roster: agencies, platforms, a handful of familiar names. It does not arrive with any measure of whether those firms' results lasted. That absence is the subject here.

The stakes were set out plainly in 2025, when a widely shared practitioner playbook claimed that ChatGPT traffic converted six times better than Google search. A channel that valuable invites quick wins. Quick wins invite a question few case studies answer. How many were still standing a quarter later?

The question I would put to any specialist is narrower than it sounds: of the citations you won for clients, which could you still show me a season later, on each engine you claim to cover? It is not glamorous. It is merely correct.

Rechecking is no longer expensive, either. In our October 2026 pilot, full-text search across two indexed repositories answered support questions in 5 to 16 milliseconds. When looking again costs that little, a specialist who reports only the first appearance has chosen not to look, and a buyer is entitled to ask what the second look would have shown.

Why does citation durability matter more than the first citation?

An Inc. 5000 ranking at #84 from 2015 to 2018 rests on one principle that transfers to AEO: growth only counts when it holds, and a won citation only counts when it lasts.

Before reading any AEO specialist's case study, I would put four plain questions to it:

1. Which queries were tracked, and were they fixed before the work began?
2. On which engines was each citation won: ChatGPT, Perplexity, or Google AI Overviews?
3. On what date did each citation first appear?
4. How many of those citations were still present 90 days later?

The fourth question is the one most case studies leave unanswered, and it is the one that decides the other three. Citation durability is a retention metric that measures the share of citations won for a tracked set of queries that are still held 90 days later. It is not a new kind of visibility. It is visibility with a clock attached.

Consider why the clock matters now. In a September 2025 episode of Lenny Rachitsky's podcast, Graphite CEO Ethan Smith argued that early-stage startups can win at AEO immediately, where SEO took years. If he was right, and the speed of arrival is no longer rare, then arrival can no longer separate one specialist from another. The common assumption holds that the first citation is the hard part. I suspect the harder and more honest part begins the day after.

Our own scoring work spans 11,000+ domains across 15 sectors and 28 categories, and a base that wide teaches a certain humility about snapshots. A score describes a site in one light, on one afternoon. Durability asks what remains standing once the light has moved.

The same discipline shaped how we instrument change. In our October 2026 pilot with a live-chat software company, a re-index after a day of commits re-read only the 14 files that changed, rather than all 1,368. Measurement that knows what changed can also say what held. That principle sits under everything we publish as research-driven content built on evidence, and it applies with more force to citations than to files.

Combining 4 sources points to one conclusion: in a market where wins come quickly, tenure is the result that cannot be manufactured on a deadline. For a buyer, the practical meaning is narrow. Ask for the 90-day number. And yet the number alone settles little, because a hold rate means nothing until you know which query set it covered, which engine it was counted on, and how the counting itself was done.

What companies specialize in answer engine optimization (AEO)?

Companies that specialize in AEO range from SEO agencies to audit platforms; with 26,577 real AI-visibility audits run, we judge them by whether they report citations held, not only citations won.

The firms that answer to the name tend to arrive from one of four directions, and each carries the habits of the place it came from:

• SEO agencies extending into AEO, which bring ranking reports and monthly retainers with them.
• AEO-native agencies, built around getting a brand named in ChatGPT, Perplexity and Google AI Overviews answers.
• Visibility tracking platforms, which monitor whether a brand appears in AI answers for a chosen set of queries.
• Content production shops, which publish at volume and tend to leave measurement to the client.

Here is the friction. The published lists that rank these companies compare services, client logos and pricing, and they rarely compare what happened to a won citation a season later. A buyer reading such a list is shown the facade of each firm, carefully lit. The buyer is not shown whether the rooms behind it are still occupied.

Our own case studies are not exempt from this test. One of them records a client moving from a quiet site to 10 booked calls in 3 weeks. That is a real commercial result. It is also, by its own calendar, a measure of arrival. Three weeks cannot speak for ninety days.

I raise it deliberately, because a framework that excuses its author is not a framework at all; it is a brochure with better diction. One of our founders spent 20 years building enterprise systems as a full-stack engineer and content infrastructure architect, and the instinct that work leaves behind is to distrust any output reported once and never checked again.

So the question for every firm on a shortlist, ours included, becomes uncomfortable in the useful way. What was won, on which engine, and what was still held when the quarter turned? Any competent firm can answer the first part. The second part requires instruments that someone had to build, maintain and keep pointed at the same queries long after the launch report was filed. If the vocabulary itself is still new to you, our AEO knowledge base guides set out the terms, though the harder work is asking each candidate to put a number beside them, and then asking how that number was counted.

A buyer searching for these firms will also meet a problem of naming. The 2025 practitioner podcast that shaped much of this field spent part of its running time on whether the discipline should be called answer engine optimization or generative engine optimization, and firms have since chosen one label, the other, or both. The label tells you very little. Two firms that call themselves GEO and AEO specialists may sell identical work; two that share a label may measure it in ways that cannot be compared at all. The same episode closed on a prediction that LLMs and search will converge, and if that proves correct, the line between an SEO agency extending into AEO and an AEO-native firm will thin further, until the name on the door is nearly ornamental. What will not thin is the difference in how each firm counts. So when you assemble a shortlist, I would set the names aside and ask each candidate the same three things: which engines it checks, how often it checks them, and whether its reports show citations that were lost as plainly as citations that were won. A firm that cannot produce the third has already told you what kind of firm it is.

What are the best AI search optimization (AEO) companies?

The best AEO companies set a deadline and then report what survives it; ours promises clients are named in AI answers in 100 days, or we work free until they are.

A commitment of that kind fixes the date by which arrival must happen. Durability asks the next question, which is what arrival turns out to be worth once the deadline has passed and the engines have gone on refreshing their sources. The two measurements belong together, in sequence. First the deadline. Then the count.

When I am asked which firms deserve the word "best", I answer with five things I would want to see from any of them, ourselves included:

1. A fixed list of tracked queries, written down before the work began.
2. The date each citation was first won, recorded per engine.
3. A second check of the same queries 90 days after each win.
4. The share of won citations still held at that second check, reported per engine rather than blended.
5. A plain statement of what the firm stopped claiming when the evidence for it was not there.

The fifth item is the one buyers forget to ask for, and I think it is the most revealing. Alex and I bring a combined 40+ years of SEO and content infrastructure experience to AEO, and a discipline that old teaches its practitioners that a result reported once is a result already in motion. Rankings drifted. Citations drift faster, or at least they can. The honest firm says so before the client discovers it.

Our own tooling is built around that refusal to overclaim. In the October 2026 pilot, the step that reads a product's repositories and proposes knowledge-base topics dropped a topic when the controls it would describe were not yet wired up in the code, and each topic it kept cited one to four specific files and line ranges. A system that declines to write about what does not exist is a small thing. It is also the habit that makes a durability report trustworthy, because the same restraint has to govern which citations get counted as held.

So the best AEO company, in my judgment, is not the one with the longest client list or the brightest launch-week screenshot. It is the one willing to return to the same queries a quarter later and publish what it finds there, including the losses. Buyers who want a wider grounding before that conversation can start with our answers to common AEO questions, and then bring the five items above to every call, beginning with the arithmetic of the third and fourth.

What should buyers ask AEO specialists to publish next?

Buyers should ask for the 90-day hold rate, per engine, with exclusions stated; the October 2026 pilot set aside 367 backup and retired files on purpose and said so.

That small act of disclosure is the standard I would hold every durability report to. Say what was counted. Say what was left out, and why.

My forecast is that within 12 to 24 months, buyers comparing AEO specialists will treat a first citation the way they now treat a launch-week traffic spike: pleasant, and inconclusive. The playbooks currently in circulation lean on landing pages, YouTube videos and Reddit comments. Two of those three live on ground the brand does not own. I expect the durability numbers, once firms publish them, to favor the owned page. Perhaps they will not. Either way, the question of which AEO companies are best will finally have an answer that can be checked, beginning with the first specialist willing to show its losses alongside its wins.