What actually changed
For twenty years search worked one way: you typed something, you got a list, you chose. Now a large share of searches return a written answer at the top, assembled from a few sources, with the list pushed below it. Many people read the answer and stop. If your business is not inside that answer, you were not in the running at all, the click never existed to be lost.
The mechanics are different too. Traditional search matches a page to a query. An AI answer is assembled from sources the model considers reliable on the topic, then attributed. So the goal shifts from ranking a page to being the source that gets quoted, which depends on how clearly your content states facts, how consistently your business is described across the web, and whether the model can parse your pages at all.
The short version
Write content that answers real questions directly and early. Make your business unambiguous, same name, same description, same facts everywhere. Mark it up so machines can read it. Get mentioned on sites these models trust. Then measure whether you are being cited, because nobody's analytics reports this by default.
An honest note: this field is two years old and moving. Nobody has a decade of case studies, and anyone claiming a guaranteed method is selling certainty that does not exist yet. What we do know is that the fundamentals, clear content, consistent entity data, real authority, are what these systems reward, and they are worth doing regardless of how the tools evolve.
The four surfaces that matter
Each behaves differently and each needs checking separately. Being cited in one does not mean being cited in the others.
Google AI Overviews
HIGHEST VOLUMEThe written answer that now appears above the ordinary results for a large share of searches. It cites a handful of sources, and being one of them matters more than holding position three below it.
What moves it: strong conventional rankings plus content that answers the question directly in the first two sentences. Overviews are largely assembled from pages that already rank.
ChatGPT search
LARGEST AUDIENCEWhen ChatGPT browses to answer a live question it fetches pages and cites them. This is separate from what the model learned in training, and it is the part you can influence quickly.
What moves it: letting its crawler in, clear factual pages, and being mentioned on the third-party sites it tends to consult, directories, comparisons, reviews.
Perplexity
CITATION-HEAVYBuilt around citing sources for everything, so it is the easiest place to see whether your content is considered credible on a topic. Smaller audience, useful signal.
What moves it: depth and specificity. It favours pages that actually answer the question over pages that market at the reader.
Gemini and Copilot
GROWINGGoogle's assistant and Microsoft's, both drawing on their own search indexes plus their models. Gemini overlaps heavily with Overviews; Copilot leans on Bing.
What moves it: the same fundamentals, plus not neglecting Bing, most Pakistani businesses have never checked whether Bing indexes them at all.
What the work covers
Writing that gets quoted
A model extracting an answer needs a passage it can lift cleanly. Marketing prose that circles the point for four paragraphs before saying anything is unusable to it. These are the patterns that get picked up.
Answer in the first two sentences
State the answer, then explain it. A model extracting a passage takes what is near the top of the section, bury the answer and it takes someone else's.
Use the question as the heading
A heading phrased the way people ask makes the match obvious. "How much does a Shopify store cost in Pakistan?" works; "Our Pricing Philosophy" does not.
Give real numbers and ranges
Prices, durations, quantities, dates. Vague copy is unquotable, and specificity is what makes a source look authoritative on the subject.
Keep passages self-contained
Each section should make sense lifted out on its own. A paragraph that depends on the previous three cannot be quoted cleanly.
Say who wrote it and why they know
A named author with real credentials, a visible publish or update date, and sources where claims need them. All three feed how much a system trusts the page.
Cut the marketing padding
Adjectives without facts contribute nothing extractable. "Leading provider of innovative solutions" tells a model exactly as much as an empty page does.
Entity clarity: making your business legible
These systems build a picture of your business from everything they have read about it. If half the web calls you one thing and half another, if your description differs on every profile, if nobody states plainly what you do and where, the picture stays blurry, and a blurry entity does not get recommended.
One business name
Exactly the same everywhere, including punctuation. Variants read as different businesses and split whatever recognition you have earned.
One-line description
A single sentence stating what you do and where, reused verbatim across every profile so the definition never drifts.
Organization schema
Name, logo, address, contact, founder and social profiles marked up on your site so the core facts are stated rather than guessed.
Named people
A real founder and team with bios and credentials. Businesses with visible people attached are treated as more credible than anonymous ones.
sameAs links
Schema links tying your site to your LinkedIn, Facebook and other profiles, so scattered mentions resolve to one entity.
Consistent contact details
One phone number and one address across every listing. Old numbers on stale profiles quietly weaken the whole picture.
Let the right crawlers in
AI systems use their own crawlers, separate from Googlebot, and many sites block them by accident, or on purpose, following advice written for publishers protecting paid content. For a business that wants to be recommended, blocking them removes you from the answer entirely.
Usually allow
The crawlers that fetch pages to answer a live question and cite the source, OAI-SearchBot, PerplexityBot, Google-Extended for AI features. Blocking these is choosing not to appear.
Your call
Crawlers that collect content for model training rather than for citing you, like GPTBot and CCBot. Publishers often block them; most businesses gain more from being known than they lose. We explain the trade-off rather than deciding for you.
How this gets measured
This is the hardest part and where most agencies go quiet. There is no Search Console for AI answers, so measurement is partly direct testing and partly reading the traffic signals that do exist.
Prompt test results
A fixed set of customer questions re-run monthly across each surface, recording whether you are named. The most direct measure available.
Share of voice
How often you appear versus named competitors on the same question set, the number that shows whether you are gaining ground.
Assistant referral traffic
Visits arriving from chatgpt.com, perplexity.ai and similar, tracked separately in analytics. Small volumes, unusually high intent.
Impressions against clicks
In Search Console: rankings steady while clicks fall is the fingerprint of AI answers absorbing your traffic.
Cited pages
Which of your pages get quoted, so content work concentrates on the formats that are actually working.
Brand search volume
People searching your name directly. When assistants recommend you, this is often where it shows up first.
How the engagement runs
Honest timelines
Citation behaviour is less predictable than ranking. Some pages get picked up within weeks of publishing; others take months while the model's picture of your business firms up. Anyone giving you a guaranteed date is guessing.
Who this is for
Worth doing now
Businesses whose customers research before buying, services, software, education, healthcare, B2B, anything considered rather than impulse. Also anyone already ranking well who is watching clicks fall while rankings hold, which is the signature of AI answers taking the traffic.
Do the basics first
Sites that are not yet ranking in ordinary search, or that have unresolved technical problems. AI SEO builds on the same foundation, crawlable, clear, credible, so there is no shortcut that skips it. Fix technical and content first; this layers on top.
Questions we get asked
Is this actually different from normal SEO, or a new name for it?
The foundation is shared, crawlable, clear, credible content wins in both. What differs is the target and the tactics: you are optimising to be quoted rather than clicked, which changes how pages are structured, how much entity consistency matters, and how success is measured. Roughly seventy per cent overlap, thirty per cent genuinely new.
If AI answers the question, do we lose the visit?
Sometimes, and that is worth being honest about. For simple factual questions the answer often ends the journey. But for anything involving a decision, hiring, buying, comparing, being the business named in the answer is what starts the journey, and those visits convert better than ordinary search traffic because you arrived pre-recommended.
Can you guarantee ChatGPT will recommend us?
No, and nobody honestly can. There is no submission form and no ranking dashboard. What we can do is remove the reasons you are being skipped and build the signals these systems demonstrably favour, then measure the change with repeatable prompt testing.
Should we block AI crawlers to protect our content?
That advice comes from publishers who sell content and lose money when it is summarised. For a business that wants customers, blocking the citing crawlers means choosing not to appear in the answer. Training-only crawlers are a genuine judgement call and we will explain the trade-off, but we usually recommend allowing the ones that cite.
Does this replace our Google rankings work?
No. It depends on it. AI Overviews are largely built from pages that already rank, and assistants consult search results when they browse. Conventional SEO is the input, not the alternative.
How do we know it is working?
A fixed set of customer questions, re-tested monthly on every surface, recorded so you can see the change. Alongside that: referral traffic from assistants, brand search volume, and the clicks-versus-impressions pattern in Search Console.
Is this worth doing for a small local business?
Increasingly yes, because AI answers to "near me" questions draw heavily on Business Profile and review data, which means local SEO and AI visibility overlap almost entirely. If you are doing local SEO properly, you are already most of the way there.
What does it cost?
It depends on how much content restructuring is needed and how many surfaces you want tracked. The baseline visibility check is free, and it comes with a scoped quote so you see the number before deciding.
Find out if AI mentions you
Tell us what your customers would ask. We run those questions through ChatGPT, Gemini, Perplexity and Google's AI Overviews and send you what comes back, who gets named, and whether it is you. No charge.
Request sent
WhatsApp should have opened with your details ready to send. We run the questions and come back within a working day.
Related topics
Guides on AI SEO from our team
AI SEO vs Traditional SEO: What Differs
Most of the work is the same. The seventy per cent that overlaps, the thirty per cent that does not, and which one deserves your budget now.
GuideAEO and GEO: How to Get Cited by ChatGPT and AI Overviews
Answer engine optimisation explained, what AEO and GEO are, how AI search picks sources, and the page structure that gets you quoted.
GuideChatGPT SEO: How to Get Your Business Mentioned
How ChatGPT decides which businesses to name, and the seven things that make yours one of them. Written from testing against live accounts, not speculation.