AI and SEO: What We Stopped Doing, What We Kept, and What We Started

A while back, most conversations about AI and SEO in our team went one of two ways. Either someone wanted to generate fifty articles before lunch, or someone was convinced search was finished. Neither turned out to be true, and the reality has been more interesting.
Here is what has changed in our day-to-day SEO work, sorted into three piles: what we stopped, what we kept, and what we started.
What We Stopped Doing
Publishing AI drafts as they come. Raw AI text reads fine and says nothing. It repeats what the top ten results already say, which is exactly the content search engines have no reason to prefer. We now treat a first AI draft as a rough outline at best.
Writing for keywords instead of questions. Keyword density charts feel like a different era. Search is increasingly semantic, so the real question is whether a page fully answers what the person meant, not whether a phrase appears nine times.
Trusting AI with facts. Statistics, dates, policy details and product claims all get checked against the original source. We have caught confident but wrong numbers often enough that this is now a rule, not a habit.
Judging success by clicks alone. More on this below, because it changed how we report.
What We Kept
It might surprise people, but most of the basics stayed exactly where they were.
Crawlability and clean technical foundations. If a page cannot be crawled, indexed and loaded quickly, no amount of clever content helps.
Search intent matching. Informational, commercial and local queries still need different pages.
Internal linking and topical authority. Covering a subject properly across connected pages still beats a scattering of unrelated posts.
E-E-A-T. Real experience, named authors, honest sourcing and an accurate About page matter more now, not less, because generic content is everywhere.
AI search features like AI Overviews still pull from the open web. The sites they draw on are usually the ones doing the basics well.
What We Started
Using AI as an Assistant, Not an Author
Where AI genuinely saves us time is the grunt work around the writing:
Grouping hundreds of Search Console queries into topic clusters
Summarising long competitor pages so we can spot gaps faster
Turning messy client notes into a first-pass outline
Suggesting questions people might ask around a topic, which we then verify against real search data
Spotting thin or outdated sections across old posts
The thinking, the examples, the opinions and the final edit stay human.
Writing Answer-First
We now open key sections with a clear, direct answer and then add the detail. It helps readers who are skimming, and it makes pages easier for both search engines and AI assistants to understand and quote.
Checking How a Brand Shows Up in AI Assistants
We ask ChatGPT, Gemini and similar tools plain-language questions about a client's category and note who gets mentioned. It is not a perfect science, but it shows us gaps that rankings reports never will.
Refreshing Instead of Only Publishing
Updating an older post with fresh examples, a corrected fact and a better opening often does more than writing a new one. We schedule refreshes the way other teams schedule releases.
A Real Example That Changed Our Reporting
On one of our own informational pages, a query's impressions in Search Console fell by roughly 90 percent in a short window. Nothing was wrong with the page. When we searched the query ourselves, an AI Overview was answering it fully at the top, citing a handful of other sites.
That moment shifted how we think about performance. Clicks on that kind of query were never coming back, so we started tracking impressions, branded searches, direct visits and whether our content appears as a cited source. If you only watch clicks, you can misread a platform change as a quality problem.
A Simple Workflow You Can Borrow
Start with the real question behind the query, using Search Console and people's actual wording.
Use AI to map related questions and gaps, then verify them.
Write the draft yourself, adding something only you have: a client story, a mistake, a screenshot, a specific opinion.
Fact-check every claim and number against the original source.
Add clear headings, a direct answer near the top and a short FAQ where it genuinely helps.
Set a date to revisit and refresh the page.
Mistakes We See Most Often
Publishing at volume with no human edit
Copying the same structure as every other result
Ignoring author details and sources
Treating a traffic dip as a penalty before checking whether the search results page itself changed
Chasing every new acronym instead of strengthening the basics
AI has not replaced SEO. It has raised the bar for what is worth publishing, and it has made the lazy shortcuts less rewarding. If you are building content right now, the safest bet is the old one: be useful, be accurate and be clearly written by someone who knows the subject.
If you want to compare notes on how this plays out for local and service businesses, we write about it often at TechBound, a digital marketing agency in Trivandrum.