Nobody wants to hear "three to six months" about anything involving AI. The software updates weekly, the models get replaced, and it feels like results should arrive at the same speed.
They do not, and the reason is structural rather than disappointing. Half of this work is technical and moves quickly. The other half is reputational, and reputation accumulates at human speed no matter how fast the machines iterate.
How long does AEO take to work?
Structural fixes to your own pages typically show movement in about 2 to 4 weeks. Consistent citations across repeated queries take roughly 8 to 16 weeks. Visibility that holds up simultaneously across ChatGPT, Perplexity and Google AI Overviews generally takes 4 to 6 months (WebFX, AirOps).
Key takeaways
- Two speeds run in parallel: fast technical work on your own site, slow authority work everywhere else.
- First movement in about 2 to 6 weeks, stable citation patterns in 8 to 16 weeks, dependable multi engine visibility in 4 to 6 months.
- Businesses with no existing footprint sometimes move fastest, because they are starting from a blank signal rather than fighting established competitors.
- Measurement is genuinely noisy, so a short test window will mislead you in both directions.
- Being cited and being paid are different milestones, and the second one lags the first.
What actually moves in the first month?
The packaging of content you already have.
Schema markup, question shaped headings, direct answers placed at the top of sections, fixing pages that block crawlers, cleaning up the structure of your most important pages. These are changes to how existing, already indexed content is presented, which is why they land fast, in the range of 2 to 4 weeks for an established domain (WebFX).
Basic visibility and brand mention changes can begin appearing in a similar 2 to 6 week window, particularly on platforms that rely on a fresh live web index rather than a frozen training snapshot.
If you are blocking AI crawlers, unblocking them belongs in week one. It is the only change in this list that can take you from structurally impossible to merely difficult.
One thing that helps in this phase: focus. Graphite's internal analysis found roughly one in twenty landing pages drives about 85 percent of a site's traffic, which is a useful reminder that rewriting every page is a poor use of the first month. Rewrite the two or three that matter.
What takes about three months?
Consistency.
There is a large difference between appearing once in an AI answer and appearing reliably when the same question is asked in different words on different days. That second state, stable citation behavior rather than a lucky hit, tends to land in the 8 to 16 week range (WebFX).
Practitioner sources converge on roughly a 90 day window before citation share shifts meaningfully (AirOps). That is also roughly how long genuine outside mentions take to start accumulating, assuming you began that work on day one rather than after the technical phase.
What takes six months or more?
Multi engine visibility and measurable referral traffic.
Being cited across ChatGPT, Perplexity and Google AI Overviews at the same time typically takes 4 to 6 months, and AI driven referral traffic usually takes around 6 months to become a channel you can actually track and report on rather than noise in the analytics (AirOps).
This is the phase where the work stops being a project and becomes a habit, because the input is a steady stream of credible third party mentions rather than a one time deployment.
Why is it slower than "AI moves fast" suggests?
Because the strongest predictor of citation is not something you can deploy.
Ahrefs studied 75,000 brands and found that unlinked brand mentions across the web correlated with AI Overview presence at 0.664, against 0.218 for backlinks (Ahrefs). The signal that matters most is other people talking about you, and that is a reputational asset that accrues over months.
There is a second, quieter reason. When an AI answers from memory rather than searching, it is drawing on training data with a cutoff commonly twelve to eighteen months before the model shipped (Wikipedia). Work you do today cannot appear in a model that was trained last year. It can appear in live retrieval immediately, which is the faster and more controllable path, but the memory layer runs on a training cycle you have no influence over.
So the honest framing is that you are playing two games with different clocks. One is measured in weeks. One is measured in model generations.
What makes it faster or slower for a specific business?
Four factors do most of the work.
Your starting footprint. A business that already ranks, has reviews and gets written about is mostly doing packaging work, and packaging is fast. A business with no external mentions is building a signal from zero.
Competitive density. In a crowded category you are competing against reputations that took years to build. In a thin local category, the bar is often just being the business with clean structured data and a few genuine third party mentions.
Whether you are a brand new entity. Counterintuitively, this can be an advantage for speed. There is a documented case of a brand new local business being cited by both Gemini and ChatGPT within roughly a week of a structured press release going out, because there was no competing signal to displace. That advantage fades quickly once real competition exists.
Which questions you are targeting. AI prompts run much longer than search queries, often ten words or more against roughly three or four for a typed search. Long specific questions have far less competition than the short generic ones most businesses fixate on, and they move faster.
How should you measure progress without fooling yourself?
Carefully, and over more runs than feels necessary.
These systems are probabilistic. The same prompt can return different sources on different runs, and results shift with paraphrasing, interface, model version, location and whether you are logged in (Search Engine Land). An academic treatment of the problem found that measurement error only falls to an acceptable level after roughly 7 repeated runs of the same prompt (arXiv:2604.07585).
Three further honesty notes worth building into your expectations:
- There is no industry standard scoring method. Different tools define AI visibility differently, weight platforms differently and test against different prompt sets, so a score from one vendor is not comparable to another vendor's score for the same business.
- Mentions and citations are not the same thing, and they are frequently conflated in vendor reporting. Being named in an answer and being cited as a source are different outcomes.
- Your own manual checks are biased. Logged in sessions carry memory and location personalization. One practitioner described running a visibility test with account memory left on and receiving a confidently stated, entirely false claim about a well known company. A neutral session with no saved history is a better instrument than your own account.
The practical approach is to take a baseline before you change anything, then re measure the same prompts the same way at 30, 60 and 90 days. SignalCheck is built for that baseline step: it runs a live grounded query from a neutral session and shows the raw answer text, so you have a dated record of what an AI said about your category before you started, and something honest to compare against later.
When does any of this turn into money?
Later than visibility does, and not automatically.
This is the caveat the market mostly skips. Agencies working in the space have publicly noted that brands earn citations in ChatGPT and AI Overviews without a matching lift in conversions or pipeline, and that the failure is usually in the system around the citation rather than in the citation strategy.
Some of the upside numbers are real but need labels. One growth agency reports that a client, Webflow, saw a 6 times conversion rate difference between traffic arriving from AI tools and traffic arriving from Google search, attributed to AI referred visitors arriving further down the funnel with their questions already answered. That is a single agency reporting a single client's figure, not independent research, and it should be read that way.
Independently measured data does show the channel growing. Adobe Analytics, tracking more than a trillion visits to US retail sites, measured generative AI referral traffic rising 693 percent year over year across the 2025 holiday season, while noting in the same release that the base of users remains modest (Adobe). Both halves of that sentence matter. Forrester separately found 94 percent of business buyers now use AI somewhere in their buying process, up from 89 percent a year earlier (Forrester).
The reasonable expectation: visibility gains in months two and three, traffic you can attribute around month six, and revenue impact that depends entirely on whether the page an AI sends people to was built for a visitor who arrives already half convinced.
Frequently asked questions
What is the fastest thing I can do this week? Check whether your site blocks AI crawlers, and fix it if it does. Then rewrite the single most commercially important page so each section opens with a direct answer under a question shaped heading. Both are cheap and both sit in the fast lane.
Is there any point starting if I cannot commit for six months? Yes, with adjusted expectations. The technical layer is worth doing on its own merits and pays back in weeks. What you cannot compress into a short window is the authority layer, so a three month engagement that promises multi engine citation dominance is promising something the timeline data does not support.
Should I add an llms.txt file to speed things up?
It is cheap and harmless, but as of 2026 no major AI provider has committed to reading it the way Google reads a sitemap (Firecrawl). Add it if you like, but do not count it as a lever.
How often should I re check my visibility? Monthly is enough for most businesses, and more often than that mostly measures noise. What matters is running the same prompts, the same number of times, in the same neutral conditions, so the comparison means something.
Does this replace my existing SEO work? No. Answer engines can only quote content that was crawled and indexed first, so classic search hygiene remains the foundation. The honest industry read is that a meaningful share of this discipline is existing practice with new vocabulary, with one practitioner estimate putting it at roughly 40 percent genuinely new mechanics (xpert.digital). Treat AEO as an extension of what you already do, not a replacement for it.